hlt_project.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "# **SemEval 2015 Task 12**"
      ],
      "metadata": {
        "id": "aQllfvzysWsv"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "bC8devvBgwdH"
      },
      "outputs": [],
      "source": [
        "!pip install transformers\n",
        "from tokenizers import BertWordPieceTokenizer\n",
        "from transformers import BertTokenizer, TFBertModel, BertConfig\n",
        "import os\n",
        "# Save the slow pretrained tokenizer\n",
        "slow_tokenizer = BertTokenizer.from_pretrained(\"bert-base-uncased\")\n",
        "save_path = \"bert_base_uncased/\"\n",
        "if not os.path.exists(save_path):\n",
        "    os.makedirs(save_path)\n",
        "slow_tokenizer.save_pretrained(save_path)\n",
        "\n",
        "# Load the fast tokenizer from saved file\n",
        "tokenizer = BertWordPieceTokenizer(\"bert_base_uncased/vocab.txt\", lowercase=True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Pm9F7MGCr08B",
        "outputId": "5558c5d7-4cc7-4eed-b97c-d09a270e6744"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/drive\n"
          ]
        }
      ],
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        " encoder = TFBertModel.from_pretrained(\"bert-base-uncased\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 156,
          "referenced_widgets": [
            "77b9ac899d724d0eb4246ba47061eb71",
            "b1f31b25e15149b89f6e04827439b1f2",
            "fcec08fd554a4284aafa08eeba9a9fed",
            "9e5b0ae4e386465b892af86c37e21fb7",
            "05022614f2f945dfbb79c7ecda2b5cf8",
            "9d5593ff4a6a4653b23eda14d497ccff",
            "8da2cbc75ffe4c649838d0dc7a0a1990",
            "bd36a07085aa40a08adba7844b9e25f6",
            "1397738b8e794d28ad66d22f5fc2a1d1",
            "26f0dcf3c5624240983563ad63954490",
            "dc07f6b87a8d441d8ffe8d7998629e0f"
          ]
        },
        "id": "wmpTqjoOtAY2",
        "outputId": "0bb1baa1-f7be-4cd2-d536-9ff43359e2e6"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Downloading:   0%|          | 0.00/511M [00:00<?, ?B/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "77b9ac899d724d0eb4246ba47061eb71"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Some layers from the model checkpoint at bert-base-uncased were not used when initializing TFBertModel: ['nsp___cls', 'mlm___cls']\n",
            "- This IS expected if you are initializing TFBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing TFBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "All the layers of TFBertModel were initialized from the model checkpoint at bert-base-uncased.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFBertModel for predictions without further training.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "encoder(slow_tokenizer(\"hello how are you\", return_tensors=\"tf\"))['last_hidden_state'][0]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ly5z7qS0tBT-",
        "outputId": "da99b5fe-ea95-4f3d-c0f5-78d4899c3956"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor: shape=(6, 768), dtype=float32, numpy=\n",
              "array([[ 3.5377089e-02,  4.9712881e-03, -1.6398986e-01, ...,\n",
              "        -2.9469514e-01,  1.0046009e-01,  2.6906958e-01],\n",
              "       [ 1.9154608e-01,  1.5503927e-01,  3.9772588e-01, ...,\n",
              "        -6.0796332e-01,  5.2366585e-01, -2.5391939e-01],\n",
              "       [-3.5802642e-01, -6.1634368e-01,  4.8258886e-02, ...,\n",
              "        -2.8904155e-04,  7.1064633e-01, -6.0931194e-01],\n",
              "       [ 4.1431257e-01, -1.3597898e+00,  1.8618979e-01, ...,\n",
              "        -4.9402261e-01,  4.3869135e-01, -3.7817371e-01],\n",
              "       [-2.7346632e-01, -1.2541181e+00,  3.3194888e-01, ...,\n",
              "        -5.6550704e-02,  3.8558537e-01, -5.8725363e-01],\n",
              "       [ 5.2414024e-01, -2.5613758e-01, -3.2033288e-01, ...,\n",
              "         4.1231364e-01, -3.8296321e-01, -2.1611504e-01]], dtype=float32)>"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MnQoO2eprypa"
      },
      "source": [
        "# Dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "WEWFnkYcr56Q"
      },
      "outputs": [],
      "source": [
        "import zipfile\n",
        "import pandas as pd\n",
        "import xml.etree.ElementTree as ET\n",
        "# !pip install tensorflow_text\n",
        "\n",
        "dataset_path = \"/content/drive/MyDrive/HLT/RestaurantTR.zip\"\n",
        "with zipfile.ZipFile(dataset_path,\"r\") as zip_ref:\n",
        "    zip_ref.extractall(\"./dataset\")"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Dataset creation**"
      ],
      "metadata": {
        "id": "NwM_j4P7l2-5"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "_3CTK1wOshFO"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "xml_path = \"/content/dataset/ABSA15_RestaurantsTrain/ABSA-15_Restaurants_Train_Final.xml\"\n",
        "mytree = ET.parse(xml_path)\n",
        "reviews = mytree.getroot()\n",
        "rows_list = []\n",
        "for rew in reviews:\n",
        "  rid = rew.attrib['rid']\n",
        "  for sent in rew[0]: #cicle on single sentence in review.\n",
        "    sid = sent.attrib['id']\n",
        "    txt = sent[0].text\n",
        "    if len(sent)>1:\n",
        "      row_dict = {}\n",
        "      row_dict['rid'] = rid # Review id\n",
        "      row_dict['sid'] = sid # Sentence id\n",
        "      row_dict['txt'] = txt\n",
        "      row_dict['OOTS'] = False\n",
        "      row_dict['f'] = 'yes'\n",
        "      for opinion in sent[1]:\n",
        "        op_dict = opinion.attrib\n",
        "        rows_list.append({**row_dict,**op_dict})\n",
        "    else:\n",
        "      rows_list.append({'rid':rid,'sid':sid,'txt':txt,'OOTS':True,'target':np.nan,'category':np.nan,'polarity':np.nan,'from':0,'to':0,'f':'no'})\n",
        "df = pd.DataFrame(rows_list)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from matplotlib import pyplot as plt\n",
        "fig, axes = plt.subplots(1,2,figsize=(10,7))\n",
        "df.groupby(\"rid\").count()['sid'].plot.hist(ax=axes[0])\n",
        "df.groupby(\"rid\").count().boxplot(column=['sid'],ax=axes[1])\n",
        "axes[0].set_title(\"Sentence count distribution per review\")\n",
        "axes[1].set_title(\"Boxplot Sentence count per review\")\n",
        "axes[0].set_xlabel('number of sentences')\n",
        "axes[0].set_ylabel('frequency')\n",
        "axes[1].set_ylabel('number of sentences')\n",
        "plt.savefig(\"sentences_distribution.png\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 458
        },
        "id": "bvSqb03-zPad",
        "outputId": "46ee78aa-5092-4fa5-abb7-0dc1a15b8837"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x504 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from matplotlib import pyplot as plt\n",
        "fig, axes = plt.subplots(1,2,figsize=(10,5))\n",
        "df.groupby(\"sid\").count()['rid'].plot.hist(ax=axes[0])\n",
        "df.groupby(\"sid\").count().boxplot(column=['rid'],ax=axes[1])\n",
        "axes[0].set_title(\"Opinion count distribution per sentence\")\n",
        "axes[1].set_title(\"Boxplot opinion count per sentence\")\n",
        "axes[0].set_xlabel('number of opinions')\n",
        "axes[0].set_ylabel('frequency')\n",
        "axes[1].set_ylabel('number of opinions')\n",
        "plt.savefig(\"opinions_distribution.png\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        },
        "id": "KZvukcEL_e63",
        "outputId": "2d0ad181-5199-4713-f21a-a741dc2fe3b4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tmp = df.groupby(\"sid\").aggregate(list).reset_index()\n",
        "tmp['sentences']=tmp['target'].apply(lambda x: 'no opinions' if np.nan in x else 'has opinions')\n",
        "fig, ax = plt.subplots(figsize=(7,7))\n",
        "tmp['sentences'].value_counts().plot.pie(ax=ax, autopct=\"%.1f%%\");\n",
        "plt.title(\"Percentage of the sentences with opinions and without opinions.\")\n",
        "plt.savefig(\"sentence_wo_opinions.png\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 427
        },
        "id": "HRG7xie0B70Z",
        "outputId": "ce375ce8-fcc9-4f5f-cf4a-70f910af48c7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 504x504 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tmp = df[df['f']=='yes']\n",
        "tmp['a']=tmp['target'].apply(lambda x: 'no target' if 'NULL' in x else 'has target')\n",
        "fig, ax = plt.subplots(figsize=(7,7))\n",
        "tmp['a'].value_counts().plot.pie(ax=ax, autopct=\"%.1f%%\");\n",
        "plt.title(\"Percentage of opinions with target.\")\n",
        "plt.savefig(\"target_opinions.png\")\n",
        "print(tmp.shape[0])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 548
        },
        "id": "lCb9q9LBEwcn",
        "outputId": "2bd3646c-f187-4782-8ea1-df73e449288d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:2: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame.\n",
            "Try using .loc[row_indexer,col_indexer] = value instead\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  \n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1654\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 504x504 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Cartesian product of the categories+polarities**"
      ],
      "metadata": {
        "id": "1Ei4u3cvl7Wb"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "LjQcNWNfwyLp",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "674c78f4-dd35-44ee-dbaf-9ed4e0c15de6"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "39\n"
          ]
        }
      ],
      "source": [
        "import itertools\n",
        "categories = set(df[df['category'].notna()]['category'].values)\n",
        "polarities = set(df[df['polarity'].notna()]['polarity'].values)\n",
        "cat_pol_couples = set(list(itertools.product(categories,polarities)))\n",
        "print(len(cat_pol_couples))"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "cat_pol_couples"
      ],
      "metadata": {
        "id": "TOSOHqCpqSG2",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "af6203fa-8d97-4f44-f11a-6e96980a9ec5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{('AMBIENCE#GENERAL', 'negative'),\n",
              " ('AMBIENCE#GENERAL', 'neutral'),\n",
              " ('AMBIENCE#GENERAL', 'positive'),\n",
              " ('DRINKS#PRICES', 'negative'),\n",
              " ('DRINKS#PRICES', 'neutral'),\n",
              " ('DRINKS#PRICES', 'positive'),\n",
              " ('DRINKS#QUALITY', 'negative'),\n",
              " ('DRINKS#QUALITY', 'neutral'),\n",
              " ('DRINKS#QUALITY', 'positive'),\n",
              " ('DRINKS#STYLE_OPTIONS', 'negative'),\n",
              " ('DRINKS#STYLE_OPTIONS', 'neutral'),\n",
              " ('DRINKS#STYLE_OPTIONS', 'positive'),\n",
              " ('FOOD#GENERAL', 'negative'),\n",
              " ('FOOD#GENERAL', 'neutral'),\n",
              " ('FOOD#GENERAL', 'positive'),\n",
              " ('FOOD#PRICES', 'negative'),\n",
              " ('FOOD#PRICES', 'neutral'),\n",
              " ('FOOD#PRICES', 'positive'),\n",
              " ('FOOD#QUALITY', 'negative'),\n",
              " ('FOOD#QUALITY', 'neutral'),\n",
              " ('FOOD#QUALITY', 'positive'),\n",
              " ('FOOD#STYLE_OPTIONS', 'negative'),\n",
              " ('FOOD#STYLE_OPTIONS', 'neutral'),\n",
              " ('FOOD#STYLE_OPTIONS', 'positive'),\n",
              " ('LOCATION#GENERAL', 'negative'),\n",
              " ('LOCATION#GENERAL', 'neutral'),\n",
              " ('LOCATION#GENERAL', 'positive'),\n",
              " ('RESTAURANT#GENERAL', 'negative'),\n",
              " ('RESTAURANT#GENERAL', 'neutral'),\n",
              " ('RESTAURANT#GENERAL', 'positive'),\n",
              " ('RESTAURANT#MISCELLANEOUS', 'negative'),\n",
              " ('RESTAURANT#MISCELLANEOUS', 'neutral'),\n",
              " ('RESTAURANT#MISCELLANEOUS', 'positive'),\n",
              " ('RESTAURANT#PRICES', 'negative'),\n",
              " ('RESTAURANT#PRICES', 'neutral'),\n",
              " ('RESTAURANT#PRICES', 'positive'),\n",
              " ('SERVICE#GENERAL', 'negative'),\n",
              " ('SERVICE#GENERAL', 'neutral'),\n",
              " ('SERVICE#GENERAL', 'positive')}"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Dataset augmentation with all possible combination of category and polarities different from the one with the values.**"
      ],
      "metadata": {
        "id": "sYLHcJAXmM0K"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Mi_-38uiF8Jw"
      },
      "outputs": [],
      "source": [
        "# Dataset augmentation with all possible combination of category and polarities different from the one with the values.\n",
        "grouped_by_sid = df.groupby('sid').agg(list).reset_index()\n",
        "new_rows = []\n",
        "for index, row in grouped_by_sid.iterrows():\n",
        "  cats = row['category']\n",
        "  pols = row['polarity']\n",
        "  current_list = set()\n",
        "  if len(cats)>0 and (np.nan not in cats):\n",
        "    for (i,c) in enumerate(cats):\n",
        "      current_list.add((cats[i],pols[i]))\n",
        "  for (newc,newpol) in cat_pol_couples.difference(current_list): # nuovi da aggiungere\n",
        "    new_row = {'rid':row['rid'][0],'sid':row['sid'],'txt':row['txt'][0],'OOTS':'False','target':np.nan,'category':newc,'polarity':newpol,'from':0,'to':0,'f':'no'}\n",
        "    new_rows.append(new_row)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "aIY9lF3qMhpT",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "outputId": "1290fb8e-af70-4eff-b39d-9989f0190ac3"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "           rid        sid                                                txt  \\\n",
              "0      1004293  1004293:0  Judging from previous posts this used to be a ...   \n",
              "1      1004293  1004293:1  We, there were four of us, arrived at noon - t...   \n",
              "2      1004293  1004293:2  They never brought us complimentary noodles, i...   \n",
              "3      1004293  1004293:3  The food was lousy - too sweet or too salty an...   \n",
              "4      1004293  1004293:3  The food was lousy - too sweet or too salty an...   \n",
              "...        ...        ...                                                ...   \n",
              "49775   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49776   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49777   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49778   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49779   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "\n",
              "        OOTS    f    target                  category  polarity from  to  \n",
              "0      False  yes     place        RESTAURANT#GENERAL  negative   51  56  \n",
              "1      False  yes     staff           SERVICE#GENERAL  negative   75  80  \n",
              "2      False  yes      NULL           SERVICE#GENERAL  negative    0   0  \n",
              "3      False  yes      food              FOOD#QUALITY  negative    4   8  \n",
              "4      False  yes  portions        FOOD#STYLE_OPTIONS  negative   52  60  \n",
              "...      ...  ...       ...                       ...       ...  ...  ..  \n",
              "49775  False   no       NaN         RESTAURANT#PRICES   neutral    0   0  \n",
              "49776  False   no       NaN            DRINKS#QUALITY  negative    0   0  \n",
              "49777  False   no       NaN  RESTAURANT#MISCELLANEOUS  positive    0   0  \n",
              "49778  False   no       NaN           SERVICE#GENERAL  negative    0   0  \n",
              "49779  False   no       NaN  RESTAURANT#MISCELLANEOUS   neutral    0   0  \n",
              "\n",
              "[51629 rows x 10 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-25de264c-5bc9-4b1f-825c-c8108bb97b74\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>rid</th>\n",
              "      <th>sid</th>\n",
              "      <th>txt</th>\n",
              "      <th>OOTS</th>\n",
              "      <th>f</th>\n",
              "      <th>target</th>\n",
              "      <th>category</th>\n",
              "      <th>polarity</th>\n",
              "      <th>from</th>\n",
              "      <th>to</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:0</td>\n",
              "      <td>Judging from previous posts this used to be a ...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>place</td>\n",
              "      <td>RESTAURANT#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>51</td>\n",
              "      <td>56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:1</td>\n",
              "      <td>We, there were four of us, arrived at noon - t...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>staff</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>75</td>\n",
              "      <td>80</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:2</td>\n",
              "      <td>They never brought us complimentary noodles, i...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>NULL</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:3</td>\n",
              "      <td>The food was lousy - too sweet or too salty an...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>food</td>\n",
              "      <td>FOOD#QUALITY</td>\n",
              "      <td>negative</td>\n",
              "      <td>4</td>\n",
              "      <td>8</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:3</td>\n",
              "      <td>The food was lousy - too sweet or too salty an...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>portions</td>\n",
              "      <td>FOOD#STYLE_OPTIONS</td>\n",
              "      <td>negative</td>\n",
              "      <td>52</td>\n",
              "      <td>60</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49775</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#PRICES</td>\n",
              "      <td>neutral</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49776</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>DRINKS#QUALITY</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49777</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#MISCELLANEOUS</td>\n",
              "      <td>positive</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49778</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49779</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#MISCELLANEOUS</td>\n",
              "      <td>neutral</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>51629 rows × 10 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-25de264c-5bc9-4b1f-825c-c8108bb97b74')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-25de264c-5bc9-4b1f-825c-c8108bb97b74 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-25de264c-5bc9-4b1f-825c-c8108bb97b74');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 8
        }
      ],
      "source": [
        "augmented_dataset_part = pd.DataFrame(new_rows)\n",
        "df = pd.concat([df,augmented_dataset_part])\n",
        "df"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df = df[df['category'].notna()] #this is needed because we already have inserted all the category combinatios also for no opinions reviews.\n",
        "df"
      ],
      "metadata": {
        "id": "Jh68_g2FkMRn",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "outputId": "2d023a74-280f-4052-8159-a9e0888d80eb"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "           rid        sid                                                txt  \\\n",
              "0      1004293  1004293:0  Judging from previous posts this used to be a ...   \n",
              "1      1004293  1004293:1  We, there were four of us, arrived at noon - t...   \n",
              "2      1004293  1004293:2  They never brought us complimentary noodles, i...   \n",
              "3      1004293  1004293:3  The food was lousy - too sweet or too salty an...   \n",
              "4      1004293  1004293:3  The food was lousy - too sweet or too salty an...   \n",
              "...        ...        ...                                                ...   \n",
              "49775   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49776   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49777   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49778   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49779   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "\n",
              "        OOTS    f    target                  category  polarity from  to  \n",
              "0      False  yes     place        RESTAURANT#GENERAL  negative   51  56  \n",
              "1      False  yes     staff           SERVICE#GENERAL  negative   75  80  \n",
              "2      False  yes      NULL           SERVICE#GENERAL  negative    0   0  \n",
              "3      False  yes      food              FOOD#QUALITY  negative    4   8  \n",
              "4      False  yes  portions        FOOD#STYLE_OPTIONS  negative   52  60  \n",
              "...      ...  ...       ...                       ...       ...  ...  ..  \n",
              "49775  False   no       NaN         RESTAURANT#PRICES   neutral    0   0  \n",
              "49776  False   no       NaN            DRINKS#QUALITY  negative    0   0  \n",
              "49777  False   no       NaN  RESTAURANT#MISCELLANEOUS  positive    0   0  \n",
              "49778  False   no       NaN           SERVICE#GENERAL  negative    0   0  \n",
              "49779  False   no       NaN  RESTAURANT#MISCELLANEOUS   neutral    0   0  \n",
              "\n",
              "[51434 rows x 10 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-f4959b1b-9e72-49f4-81a2-65a9cfd4f80b\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>rid</th>\n",
              "      <th>sid</th>\n",
              "      <th>txt</th>\n",
              "      <th>OOTS</th>\n",
              "      <th>f</th>\n",
              "      <th>target</th>\n",
              "      <th>category</th>\n",
              "      <th>polarity</th>\n",
              "      <th>from</th>\n",
              "      <th>to</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:0</td>\n",
              "      <td>Judging from previous posts this used to be a ...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>place</td>\n",
              "      <td>RESTAURANT#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>51</td>\n",
              "      <td>56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:1</td>\n",
              "      <td>We, there were four of us, arrived at noon - t...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>staff</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>75</td>\n",
              "      <td>80</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:2</td>\n",
              "      <td>They never brought us complimentary noodles, i...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>NULL</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:3</td>\n",
              "      <td>The food was lousy - too sweet or too salty an...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>food</td>\n",
              "      <td>FOOD#QUALITY</td>\n",
              "      <td>negative</td>\n",
              "      <td>4</td>\n",
              "      <td>8</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:3</td>\n",
              "      <td>The food was lousy - too sweet or too salty an...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>portions</td>\n",
              "      <td>FOOD#STYLE_OPTIONS</td>\n",
              "      <td>negative</td>\n",
              "      <td>52</td>\n",
              "      <td>60</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49775</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#PRICES</td>\n",
              "      <td>neutral</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49776</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>DRINKS#QUALITY</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49777</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#MISCELLANEOUS</td>\n",
              "      <td>positive</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49778</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49779</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#MISCELLANEOUS</td>\n",
              "      <td>neutral</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>51434 rows × 10 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-f4959b1b-9e72-49f4-81a2-65a9cfd4f80b')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-f4959b1b-9e72-49f4-81a2-65a9cfd4f80b button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-f4959b1b-9e72-49f4-81a2-65a9cfd4f80b');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Distribution of opinions over sentences**"
      ],
      "metadata": {
        "id": "rBRg5D0VmdPd"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df['f'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "0IXa_vD3EvLT",
        "outputId": "fe948ea2-1352-40f9-afba-bc3d0701fd5b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "no     49780\n",
              "yes     1654\n",
              "Name: f, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df.groupby('sid').agg(len).reset_index()['category'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "l8B_lHnPFSaI",
        "outputId": "441ea91b-9a11-4204-82b8-a1b5c1fbb4d0"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "39    1205\n",
              "40      84\n",
              "41      18\n",
              "42       5\n",
              "43       2\n",
              "45       1\n",
              "Name: category, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Special character identification**"
      ],
      "metadata": {
        "id": "Lo74Hrgsmxy8"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sSYLYJaRl727",
        "outputId": "0ae9b086-df55-44b5-ce2b-a8d455e438a5"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "['this', 'is', 'a', 'different', '##ia', '##tor', '.', ',', '?', '!', ':']\n",
            "{'input_ids': [101, 2023, 2003, 1037, 2367, 2401, 4263, 1012, 1010, 1029, 999, 1024, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}\n"
          ]
        }
      ],
      "source": [
        "#special_caracters = [1012, 1010, 1029, 999, 1024] #in order .,?!:\n",
        "special_caracters = ['.',',','?','!'] #in order .,?!:\n",
        "intersection = lambda x: list(set.intersection(set(x),set(special_caracters)))\n",
        "word = \"this is a differentiator.,?!:\"\n",
        "print(slow_tokenizer.tokenize(word))\n",
        "print(slow_tokenizer(word))"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "t=\"ao,\"\n",
        "len(t.replace(\"#\",\"\"))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "be1uEhxPzXCe",
        "outputId": "1896d674-602b-4dea-c559-94cab679b9c3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "3"
            ]
          },
          "metadata": {},
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**New algorithm for BIO tagging**"
      ],
      "metadata": {
        "id": "4Pxf4JhzCDNE"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import re\n",
        "cleanString = lambda x: re.sub('\\W+','',x)\n",
        "list_of_tagging_list = []\n",
        "\n",
        "for index, row in df.iterrows():\n",
        "  tagging_list = []\n",
        "  if row['f'] == 'yes':\n",
        "    txt = row['txt'].split(' ')        # We split the sentence in words with ' ' spaces.\n",
        "    targets = row['target'].split(' ') # Target:'food pasta pizza'.\n",
        "    from_ = int(row['from'])\n",
        "    to = int(row['to']) \n",
        "    position = 0\n",
        "    old_tag = 'O'\n",
        "    for w in txt:\n",
        "      tokens = slow_tokenizer.tokenize(w) # We get the token for each word for istance: differentiator, -> ['different', '##ia', '##tor',',']\n",
        "      for t in tokens: #for each token\n",
        "        position+=len(t.replace(\"#\",\"\"))\n",
        "        if position-len(t.replace(\"#\",\"\"))>=from_ and position<=to:\n",
        "          if old_tag == 'O':\n",
        "            tagging_list.extend(['B'])\n",
        "          elif old_tag == 'B':\n",
        "            tagging_list.extend(['I'])\n",
        "          else:\n",
        "            tagging_list.extend(['I'])\n",
        "        else:\n",
        "          tagging_list.extend(['O'])\n",
        "      position+=1 #space\n",
        "      old_tag = tagging_list[-1]\n",
        "    token_all_phrase = slow_tokenizer.tokenize(row['txt'])\n",
        "    # print(\"txt: {} targets:{} tokens:{} tagging: {} correct: {}\".format(txt,targets,token_all_phrase,tagging_list,len(token_all_phrase)==len(tagging_list))) \n",
        "  else:\n",
        "    len_tokens = len(slow_tokenizer.tokenize(row['txt'])) # tags all O\n",
        "    tagging_list.extend(['O']*len_tokens)\n",
        "      \n",
        "  list_of_tagging_list.append(tagging_list)\n",
        "df['tagging'] = list_of_tagging_list"
      ],
      "metadata": {
        "id": "yFfGlg594WY3",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "046134d1-7b4a-418b-d57e-7c2826c2c4f7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:36: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame.\n",
            "Try using .loc[row_indexer,col_indexer] = value instead\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Check if there are some opinions with Target but without any 'B' in the tagging.\n",
        "tmp = df[df['target'].notna()]\n",
        "tmp = tmp[tmp['target']!=\"NULL\"]\n",
        "sum=0\n",
        "for i,row in tmp.iterrows():\n",
        "  if 'B' in row['tagging']:\n",
        "    sum+=1\n",
        "  else:\n",
        "    print(\"txt:{} tagging:{}\".format(row[\"txt\"],row['tagging']))\n",
        "print(sum)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "N0QU1yntzMw6",
        "outputId": "0175d266-11a8-4c50-c8f5-6453aa2a0ece"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1279\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**New column with category+polarity for the input.**"
      ],
      "metadata": {
        "id": "NvRhr6UqnKfl"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#add the category column with the polarity.\n",
        "list_aspects=[]\n",
        "for i,row in df.iterrows():\n",
        "  categ=None\n",
        "  categ = (row[\"category\"].lower()).split('#')\n",
        "  new_text = categ[0]+\" \"+categ[1]+\" \"+row['polarity']\n",
        "  list_aspects.append(new_text)\n",
        "df['categ_txt'] = list_aspects\n",
        "df\n"
      ],
      "metadata": {
        "id": "V6MTEn9k5Nd9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 750
        },
        "outputId": "5978d634-77f7-440b-b6b1-fc77124f789c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:8: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame.\n",
            "Try using .loc[row_indexer,col_indexer] = value instead\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  \n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "           rid        sid                                                txt  \\\n",
              "0      1004293  1004293:0  Judging from previous posts this used to be a ...   \n",
              "1      1004293  1004293:1  We, there were four of us, arrived at noon - t...   \n",
              "2      1004293  1004293:2  They never brought us complimentary noodles, i...   \n",
              "3      1004293  1004293:3  The food was lousy - too sweet or too salty an...   \n",
              "4      1004293  1004293:3  The food was lousy - too sweet or too salty an...   \n",
              "...        ...        ...                                                ...   \n",
              "49775   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49776   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49777   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49778   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "49779   958809   958809:4  We got a little tipsy from the sake but isn't ...   \n",
              "\n",
              "        OOTS    f    target              category  polarity from  to  \\\n",
              "0      False  yes     place    RESTAURANT#GENERAL  negative   51  56   \n",
              "1      False  yes     staff       SERVICE#GENERAL  negative   75  80   \n",
              "2      False  yes      NULL       SERVICE#GENERAL  negative    0   0   \n",
              "3      False  yes      food          FOOD#QUALITY  negative    4   8   \n",
              "4      False  yes  portions    FOOD#STYLE_OPTIONS  negative   52  60   \n",
              "...      ...  ...       ...                   ...       ...  ...  ..   \n",
              "49775  False   no       NaN  DRINKS#STYLE_OPTIONS  negative    0   0   \n",
              "49776  False   no       NaN       SERVICE#GENERAL   neutral    0   0   \n",
              "49777  False   no       NaN    RESTAURANT#GENERAL  positive    0   0   \n",
              "49778  False   no       NaN    RESTAURANT#GENERAL  negative    0   0   \n",
              "49779  False   no       NaN    FOOD#STYLE_OPTIONS  positive    0   0   \n",
              "\n",
              "                                                 tagging  \\\n",
              "0      [O, O, O, O, O, O, O, O, O, O, B, O, O, O, O, ...   \n",
              "1      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "2      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "3       [O, B, O, O, O, O, O, O, O, O, O, O, O, O, O, O]   \n",
              "4       [O, O, O, O, O, O, O, O, O, O, O, O, O, B, O, O]   \n",
              "...                                                  ...   \n",
              "49775  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "49776  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "49777  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "49778  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "49779  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "\n",
              "                           categ_txt  \n",
              "0        restaurant general negative  \n",
              "1           service general negative  \n",
              "2           service general negative  \n",
              "3              food quality negative  \n",
              "4        food style_options negative  \n",
              "...                              ...  \n",
              "49775  drinks style_options negative  \n",
              "49776        service general neutral  \n",
              "49777    restaurant general positive  \n",
              "49778    restaurant general negative  \n",
              "49779    food style_options positive  \n",
              "\n",
              "[51434 rows x 12 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-df23b423-efe7-42db-a746-8692143db9e8\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>rid</th>\n",
              "      <th>sid</th>\n",
              "      <th>txt</th>\n",
              "      <th>OOTS</th>\n",
              "      <th>f</th>\n",
              "      <th>target</th>\n",
              "      <th>category</th>\n",
              "      <th>polarity</th>\n",
              "      <th>from</th>\n",
              "      <th>to</th>\n",
              "      <th>tagging</th>\n",
              "      <th>categ_txt</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:0</td>\n",
              "      <td>Judging from previous posts this used to be a ...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>place</td>\n",
              "      <td>RESTAURANT#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>51</td>\n",
              "      <td>56</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, B, O, O, O, O, ...</td>\n",
              "      <td>restaurant general negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:1</td>\n",
              "      <td>We, there were four of us, arrived at noon - t...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>staff</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>75</td>\n",
              "      <td>80</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>service general negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:2</td>\n",
              "      <td>They never brought us complimentary noodles, i...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>NULL</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>service general negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:3</td>\n",
              "      <td>The food was lousy - too sweet or too salty an...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>food</td>\n",
              "      <td>FOOD#QUALITY</td>\n",
              "      <td>negative</td>\n",
              "      <td>4</td>\n",
              "      <td>8</td>\n",
              "      <td>[O, B, O, O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "      <td>food quality negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1004293</td>\n",
              "      <td>1004293:3</td>\n",
              "      <td>The food was lousy - too sweet or too salty an...</td>\n",
              "      <td>False</td>\n",
              "      <td>yes</td>\n",
              "      <td>portions</td>\n",
              "      <td>FOOD#STYLE_OPTIONS</td>\n",
              "      <td>negative</td>\n",
              "      <td>52</td>\n",
              "      <td>60</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, B, O, O]</td>\n",
              "      <td>food style_options negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49775</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>DRINKS#STYLE_OPTIONS</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>drinks style_options negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49776</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>SERVICE#GENERAL</td>\n",
              "      <td>neutral</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>service general neutral</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49777</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#GENERAL</td>\n",
              "      <td>positive</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>restaurant general positive</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49778</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>RESTAURANT#GENERAL</td>\n",
              "      <td>negative</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>restaurant general negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49779</th>\n",
              "      <td>958809</td>\n",
              "      <td>958809:4</td>\n",
              "      <td>We got a little tipsy from the sake but isn't ...</td>\n",
              "      <td>False</td>\n",
              "      <td>no</td>\n",
              "      <td>NaN</td>\n",
              "      <td>FOOD#STYLE_OPTIONS</td>\n",
              "      <td>positive</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>food style_options positive</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>51434 rows × 12 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-df23b423-efe7-42db-a746-8692143db9e8')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-df23b423-efe7-42db-a746-8692143db9e8 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-df23b423-efe7-42db-a746-8692143db9e8');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def merge_tagging(tags_list): # For each (text, category) the targets do not overlapp\n",
        "  result = []\n",
        "  if len(tags_list)==1:\n",
        "    return tags_list[0]\n",
        "  for i in range(len(tags_list[0])):\n",
        "    for t in tags_list:\n",
        "      val=\"O\"\n",
        "      if t[i] == \"B\" or t[i] == \"I\":\n",
        "        val = t[i]\n",
        "        break\n",
        "      else:\n",
        "        val = \"O\"\n",
        "    result.append(val)\n",
        "  return result"
      ],
      "metadata": {
        "id": "HBMldimgHcuE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "tagging_frame = df.groupby([\"txt\", \"categ_txt\"]).agg(list).reset_index()[['txt','categ_txt','f','tagging']]\n",
        "tagging_frame['tagging']= tagging_frame['tagging'].apply(lambda x: merge_tagging(x))\n",
        "tagging_frame['f']= tagging_frame['f'].apply(lambda x: \"yes\" if \"yes\" in x else \"no\")\n",
        "tagging_frame"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 423
        },
        "id": "cVyvi10npLv_",
        "outputId": "dfba99f4-83fc-4c0d-c64e-f58f4e5c032d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                   txt  \\\n",
              "0      $20 for all you can eat sushi cannot be beaten.   \n",
              "1      $20 for all you can eat sushi cannot be beaten.   \n",
              "2      $20 for all you can eat sushi cannot be beaten.   \n",
              "3      $20 for all you can eat sushi cannot be beaten.   \n",
              "4      $20 for all you can eat sushi cannot be beaten.   \n",
              "...                                                ...   \n",
              "51202                   wont come back again for sure!   \n",
              "51203                   wont come back again for sure!   \n",
              "51204                   wont come back again for sure!   \n",
              "51205                   wont come back again for sure!   \n",
              "51206                   wont come back again for sure!   \n",
              "\n",
              "                        categ_txt   f                                  tagging  \n",
              "0       ambience general negative  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "1        ambience general neutral  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "2       ambience general positive  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "3          drinks prices negative  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "4           drinks prices neutral  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "...                           ...  ..                                      ...  \n",
              "51202   restaurant prices neutral  no                 [O, O, O, O, O, O, O, O]  \n",
              "51203  restaurant prices positive  no                 [O, O, O, O, O, O, O, O]  \n",
              "51204    service general negative  no                 [O, O, O, O, O, O, O, O]  \n",
              "51205     service general neutral  no                 [O, O, O, O, O, O, O, O]  \n",
              "51206    service general positive  no                 [O, O, O, O, O, O, O, O]  \n",
              "\n",
              "[51207 rows x 4 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-81ea8cd1-1ce5-4814-861e-2f3e3855e9f3\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>txt</th>\n",
              "      <th>categ_txt</th>\n",
              "      <th>f</th>\n",
              "      <th>tagging</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>ambience general negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>ambience general neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>ambience general positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>drinks prices negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>drinks prices neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51202</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>restaurant prices neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51203</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>restaurant prices positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51204</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>service general negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51205</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>service general neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51206</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>service general positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>51207 rows × 4 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-81ea8cd1-1ce5-4814-861e-2f3e3855e9f3')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-81ea8cd1-1ce5-4814-861e-2f3e3855e9f3 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-81ea8cd1-1ce5-4814-861e-2f3e3855e9f3');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tagging_frame[tagging_frame['txt']==\"Good spreads, great beverage selections and bagels really tasty.\"]"
      ],
      "metadata": {
        "id": "O3mLhaFF7VbH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#Number of rows with a target in the tagging sequence.\n",
        "sum=0\n",
        "for i,row in tagging_frame.iterrows():\n",
        "  if 'B' in row['tagging']:\n",
        "    sum+=1\n",
        "print(sum)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "AxJhRzdAwzLN",
        "outputId": "a8220e70-54af-48a2-8f34-c8391dc57422"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1137\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tagging_frame[tagging_frame['f']==\"yes\"]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 423
        },
        "id": "dwdfHv9O-M8Y",
        "outputId": "af518ea9-31a9-4896-fb9c-179256ee6d91"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                     txt  \\\n",
              "17       $20 for all you can eat sushi cannot be beaten.   \n",
              "56     $6 and there is much tasty food, all of it fre...   \n",
              "59     $6 and there is much tasty food, all of it fre...   \n",
              "62     $6 and there is much tasty food, all of it fre...   \n",
              "164    (Always ask the bartender for the SEASONAL bee...   \n",
              "...                                                  ...   \n",
              "50994  too large for just two people but nothing was ...   \n",
              "51041                              we love th pink pony.   \n",
              "51117  well, i didn't find it there, and trust, i hav...   \n",
              "51131      whoever the jazz duo was, they were on POINT.   \n",
              "51195                     wont come back again for sure!   \n",
              "\n",
              "                         categ_txt    f  \\\n",
              "17            food prices positive  yes   \n",
              "56            food prices positive  yes   \n",
              "59           food quality positive  yes   \n",
              "62     food style_options positive  yes   \n",
              "164        drinks quality positive  yes   \n",
              "...                            ...  ...   \n",
              "50994  food style_options negative  yes   \n",
              "51041  restaurant general positive  yes   \n",
              "51117  restaurant general negative  yes   \n",
              "51131    ambience general positive  yes   \n",
              "51195  restaurant general negative  yes   \n",
              "\n",
              "                                                 tagging  \n",
              "17               [O, O, O, B, I, I, I, I, I, O, O, O, O]  \n",
              "56     [O, O, O, O, O, O, O, O, B, O, O, O, O, O, O, ...  \n",
              "59     [O, O, O, O, O, O, O, O, B, O, O, O, O, O, O, ...  \n",
              "62     [O, O, O, O, O, O, O, O, B, O, O, O, O, O, O, ...  \n",
              "164                 [O, O, O, O, O, O, O, B, I, O, O, O]  \n",
              "...                                                  ...  \n",
              "50994                  [O, O, O, O, O, O, O, O, O, O, O]  \n",
              "51041                                 [O, O, O, B, I, O]  \n",
              "51117  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "51131                  [O, O, B, I, O, O, O, O, O, O, O]  \n",
              "51195                           [O, O, O, O, O, O, O, O]  \n",
              "\n",
              "[1504 rows x 4 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-868b682c-b900-4f4b-a497-919402b26ce5\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>txt</th>\n",
              "      <th>categ_txt</th>\n",
              "      <th>f</th>\n",
              "      <th>tagging</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>17</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>food prices positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, B, I, I, I, I, I, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>56</th>\n",
              "      <td>$6 and there is much tasty food, all of it fre...</td>\n",
              "      <td>food prices positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, B, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>59</th>\n",
              "      <td>$6 and there is much tasty food, all of it fre...</td>\n",
              "      <td>food quality positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, B, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>62</th>\n",
              "      <td>$6 and there is much tasty food, all of it fre...</td>\n",
              "      <td>food style_options positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, B, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>164</th>\n",
              "      <td>(Always ask the bartender for the SEASONAL bee...</td>\n",
              "      <td>drinks quality positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, B, I, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50994</th>\n",
              "      <td>too large for just two people but nothing was ...</td>\n",
              "      <td>food style_options negative</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51041</th>\n",
              "      <td>we love th pink pony.</td>\n",
              "      <td>restaurant general positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, B, I, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51117</th>\n",
              "      <td>well, i didn't find it there, and trust, i hav...</td>\n",
              "      <td>restaurant general negative</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51131</th>\n",
              "      <td>whoever the jazz duo was, they were on POINT.</td>\n",
              "      <td>ambience general positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, B, I, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51195</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>restaurant general negative</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>1504 rows × 4 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-868b682c-b900-4f4b-a497-919402b26ce5')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-868b682c-b900-4f4b-a497-919402b26ce5 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-868b682c-b900-4f4b-a497-919402b26ce5');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 29
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Preparing dataset for the model**"
      ],
      "metadata": {
        "id": "iLsvwoojnm3K"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df = tagging_frame\n",
        "df"
      ],
      "metadata": {
        "id": "0K-8EaL2mRIq",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 423
        },
        "outputId": "d5b6aab5-8f00-4e79-e50f-a9a0fe5b1125"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                   txt  \\\n",
              "0      $20 for all you can eat sushi cannot be beaten.   \n",
              "1      $20 for all you can eat sushi cannot be beaten.   \n",
              "2      $20 for all you can eat sushi cannot be beaten.   \n",
              "3      $20 for all you can eat sushi cannot be beaten.   \n",
              "4      $20 for all you can eat sushi cannot be beaten.   \n",
              "...                                                ...   \n",
              "51202                   wont come back again for sure!   \n",
              "51203                   wont come back again for sure!   \n",
              "51204                   wont come back again for sure!   \n",
              "51205                   wont come back again for sure!   \n",
              "51206                   wont come back again for sure!   \n",
              "\n",
              "                        categ_txt   f                                  tagging  \n",
              "0       ambience general negative  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "1        ambience general neutral  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "2       ambience general positive  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "3          drinks prices negative  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "4           drinks prices neutral  no  [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "...                           ...  ..                                      ...  \n",
              "51202   restaurant prices neutral  no                 [O, O, O, O, O, O, O, O]  \n",
              "51203  restaurant prices positive  no                 [O, O, O, O, O, O, O, O]  \n",
              "51204    service general negative  no                 [O, O, O, O, O, O, O, O]  \n",
              "51205     service general neutral  no                 [O, O, O, O, O, O, O, O]  \n",
              "51206    service general positive  no                 [O, O, O, O, O, O, O, O]  \n",
              "\n",
              "[51207 rows x 4 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-5edc3242-0918-4e07-b007-7e7f97f215db\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>txt</th>\n",
              "      <th>categ_txt</th>\n",
              "      <th>f</th>\n",
              "      <th>tagging</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>ambience general negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>ambience general neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>ambience general positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>drinks prices negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>$20 for all you can eat sushi cannot be beaten.</td>\n",
              "      <td>drinks prices neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51202</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>restaurant prices neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51203</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>restaurant prices positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51204</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>service general negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51205</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>service general neutral</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>51206</th>\n",
              "      <td>wont come back again for sure!</td>\n",
              "      <td>service general positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>51207 rows × 4 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-5edc3242-0918-4e07-b007-7e7f97f215db')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-5edc3242-0918-4e07-b007-7e7f97f215db button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-5edc3242-0918-4e07-b007-7e7f97f215db');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 30
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Convering string values in integers**"
      ],
      "metadata": {
        "id": "qIaPZ_pd_ZMt"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# convert yes,no in 0,1\n",
        "# convert BIOP -> 0,1,2,3"
      ],
      "metadata": {
        "id": "C9zkovum_Y2q"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Dataset balancing**"
      ],
      "metadata": {
        "id": "BXFemE3UntMx"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.utils import shuffle\n",
        "df_yes = df[df['f']==\"yes\"]\n",
        "df_no = df[df['f']==\"no\"]\n",
        "df_yes = shuffle(df_yes)\n",
        "df_no = shuffle(df_no)\n",
        "\n",
        "counts=df_yes['categ_txt'].value_counts().values.tolist()\n",
        "cat_nums=[(v,counts[i]) for i,v in enumerate(df_yes['categ_txt'].value_counts().index.tolist())]\n",
        "df_no_balanced = pd.DataFrame([],columns=df_yes.columns)\n",
        "for c,num in cat_nums:\n",
        "  selected=df_no[df_no['categ_txt']==c]\n",
        "  selected=shuffle(selected)\n",
        "  selected=selected.iloc[:num]\n",
        "  df_no_balanced=pd.concat([df_no_balanced,selected])\n",
        "\n",
        "df=pd.concat([df_yes, df_no_balanced])\n",
        "df = shuffle(df)\n",
        "df"
      ],
      "metadata": {
        "id": "UFlGAwTxZ6hY",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 423
        },
        "outputId": "02acfde7-3131-42ce-a53e-c616d3bc649a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                     txt  \\\n",
              "48887                    fine dining restaurant quality.   \n",
              "2669   And it all comes at a very reasonable price (c...   \n",
              "26402                     Probably would not go again...   \n",
              "3234   As we were leaving, the couple standing by the...   \n",
              "7166   Everytime I decide to try another place on the...   \n",
              "...                                                  ...   \n",
              "7448                                          Fantastic!   \n",
              "28563  Steamed fresh so brought hot hot hot to your t...   \n",
              "31296        The food however, is what one might expect.   \n",
              "15317           I plan on stopping by next week as well.   \n",
              "11330          I am going to the mid town location next.   \n",
              "\n",
              "                         categ_txt    f  \\\n",
              "48887        food quality positive  yes   \n",
              "2669          food prices positive  yes   \n",
              "26402     service general positive   no   \n",
              "3234      service general negative   no   \n",
              "7166   restaurant general positive  yes   \n",
              "...                            ...  ...   \n",
              "7448      service general positive   no   \n",
              "28563         food prices negative   no   \n",
              "31296        food quality negative  yes   \n",
              "15317  restaurant general positive  yes   \n",
              "11330        food quality positive   no   \n",
              "\n",
              "                                                 tagging  \n",
              "48887                                    [O, B, O, O, O]  \n",
              "2669   [O, O, O, O, O, O, O, O, O, O, B, B, O, B, O, ...  \n",
              "26402                           [O, O, O, O, O, O, O, O]  \n",
              "3234   [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "7166   [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "...                                                  ...  \n",
              "7448                                              [O, O]  \n",
              "28563                  [O, O, O, O, O, O, O, O, O, O, O]  \n",
              "31296                     [O, B, O, O, O, O, O, O, O, O]  \n",
              "15317                     [O, O, O, O, O, O, O, O, O, O]  \n",
              "11330                     [O, O, O, O, O, O, O, O, O, O]  \n",
              "\n",
              "[3008 rows x 4 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-cbb6714a-bf81-4210-8441-59235e7f31dd\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>txt</th>\n",
              "      <th>categ_txt</th>\n",
              "      <th>f</th>\n",
              "      <th>tagging</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>48887</th>\n",
              "      <td>fine dining restaurant quality.</td>\n",
              "      <td>food quality positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, B, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2669</th>\n",
              "      <td>And it all comes at a very reasonable price (c...</td>\n",
              "      <td>food prices positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, B, B, O, B, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>26402</th>\n",
              "      <td>Probably would not go again...</td>\n",
              "      <td>service general positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3234</th>\n",
              "      <td>As we were leaving, the couple standing by the...</td>\n",
              "      <td>service general negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7166</th>\n",
              "      <td>Everytime I decide to try another place on the...</td>\n",
              "      <td>restaurant general positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7448</th>\n",
              "      <td>Fantastic!</td>\n",
              "      <td>service general positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>28563</th>\n",
              "      <td>Steamed fresh so brought hot hot hot to your t...</td>\n",
              "      <td>food prices negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>31296</th>\n",
              "      <td>The food however, is what one might expect.</td>\n",
              "      <td>food quality negative</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, B, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15317</th>\n",
              "      <td>I plan on stopping by next week as well.</td>\n",
              "      <td>restaurant general positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11330</th>\n",
              "      <td>I am going to the mid town location next.</td>\n",
              "      <td>food quality positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>3008 rows × 4 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-cbb6714a-bf81-4210-8441-59235e7f31dd')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-cbb6714a-bf81-4210-8441-59235e7f31dd button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-cbb6714a-bf81-4210-8441-59235e7f31dd');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 32
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# df.to_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")"
      ],
      "metadata": {
        "id": "lXpL8dp1df5o"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Analysis\n"
      ],
      "metadata": {
        "id": "c7bsyUUYgb3E"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from matplotlib import pyplot as plt"
      ],
      "metadata": {
        "id": "IlozPOT5nwph"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "df = pd.read_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")"
      ],
      "metadata": {
        "id": "6V8-CySOgfqT"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 337
        },
        "id": "H5TLJUwph0OH",
        "outputId": "ad982d80-a739-4245-9cac-8d52ad66586d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                     txt  \\\n",
              "12081  I choose to go with one of the special, the br...   \n",
              "47804                           You can't go wrong here.   \n",
              "31176  The food can get pricey but the prixe fixe tas...   \n",
              "49548  i've been to sapphire twice and both times the...   \n",
              "24317              Not enough wines by the glass either.   \n",
              "\n",
              "                               categ_txt    f  \\\n",
              "12081  restaurant miscellaneous negative   no   \n",
              "47804        restaurant general positive  yes   \n",
              "31176               food prices negative  yes   \n",
              "49548              food quality negative   no   \n",
              "24317              food quality positive   no   \n",
              "\n",
              "                                                 tagging  \n",
              "12081  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "47804                           [O, O, O, O, O, O, O, O]  \n",
              "31176  [O, B, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "49548  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "24317                           [O, O, O, O, O, O, O, O]  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-29f92e6b-968a-4895-ae91-6b61fc6a5b3d\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>txt</th>\n",
              "      <th>categ_txt</th>\n",
              "      <th>f</th>\n",
              "      <th>tagging</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>12081</th>\n",
              "      <td>I choose to go with one of the special, the br...</td>\n",
              "      <td>restaurant miscellaneous negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>47804</th>\n",
              "      <td>You can't go wrong here.</td>\n",
              "      <td>restaurant general positive</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>31176</th>\n",
              "      <td>The food can get pricey but the prixe fixe tas...</td>\n",
              "      <td>food prices negative</td>\n",
              "      <td>yes</td>\n",
              "      <td>[O, B, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>49548</th>\n",
              "      <td>i've been to sapphire twice and both times the...</td>\n",
              "      <td>food quality negative</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>24317</th>\n",
              "      <td>Not enough wines by the glass either.</td>\n",
              "      <td>food quality positive</td>\n",
              "      <td>no</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-29f92e6b-968a-4895-ae91-6b61fc6a5b3d')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-29f92e6b-968a-4895-ae91-6b61fc6a5b3d button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-29f92e6b-968a-4895-ae91-6b61fc6a5b3d');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 41
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df['pol'] = df.apply(lambda row: row.categ_txt.split(\" \")[-1], axis=1)\n",
        "df['tag_len'] = df.apply(lambda row: len(row.tagging), axis=1)\n",
        "df['categ'] = df.apply(lambda row: ' '.join(row.categ_txt.split(\" \")[:-1]), axis=1)"
      ],
      "metadata": {
        "id": "9QTayW8kgkTg"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "count=df[df['f']=='yes'][['categ','pol']].value_counts().reset_index()"
      ],
      "metadata": {
        "id": "yd-8uruHiU1u"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#size of the opinions with a category\n",
        "df[df['f']=='yes'].shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YcdTPlEWAZMS",
        "outputId": "f4e08515-2542-4838-ba43-c9d2da2b8340"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1504, 7)"
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import seaborn as sns\n",
        "g = sns.catplot(\n",
        "    data=count, kind=\"bar\",  x=\"categ\", y=0, hue=\"pol\", height=3, aspect=8\n",
        ")\n",
        "g.fig.set_dpi(200)\n",
        "g.despine()\n",
        "g.set_axis_labels(\"\", \"Count\")\n",
        "g.legend.set_title(\"\")\n",
        "g.fig.suptitle('Category aspect counts by polarity')\n",
        "plt.savefig('cat_counts.png')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 141
        },
        "id": "yaWiHwQnmU8k",
        "outputId": "377120d3-de7e-42b6-fb26-7a9b62e76303"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 5021.53x600 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "p=sns.displot(df, x=\"tag_len\", aspect=2)\n",
        "p.set_axis_labels(\"Number of tokens\", \"Count\")\n",
        "p.fig.suptitle(\"Opinions tokens lenght distribution\")\n",
        "plt.savefig(\"op_tokens_len.png\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 372
        },
        "id": "tMc9pkwVvWHs",
        "outputId": "662baa3b-fbc8-428e-a453-9bffdc6a5ced"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAsgAAAFjCAYAAAAgmP3GAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3de5xe8733/9cnyQhmSJCIyWSSsNGiRe1onXoKVWLvsluCksah2HcVbamflp8oquw63fW761Bnsiva2o/QBk1tdW93nKJ1jPolDpGTHBCEhhw+9x+zMnslmWSukGuuSeb1fDyux6z1Xd+11ueauR7XvOc732utyEwkSZIktehW6wIkSZKkzsSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWdIaiYjnI+JLa7tvLUTEMRHxcA3Pf15E3L4uniciMiK2LZaviYj/dy0dd2BELIiI7sX6nyPi22vj2MXx7o2IkWvreJLWTwZkqYspQuGzEfF+RLweEVdHRO9K98/MnTLzz2u779pQ68C7roqIVyNiv4+6f2b+a2ZesDbOk5mvZWZDZi75qPWUzrfSHwaZeWBm3vJxjy1p/WZAlrqQiDgduAT4IdAL2AMYBIyPiA1qWZsUET1qXYMkAZCZ6+zjq1/9akqqzNtvv5319fU5ZsyY5drffffd7NOnT95www2ZmTlq1Kj8xje+kcOHD8+Ghob8zGc+k0899VRr/0GDBuX48eNb+x522GE5YsSIbGhoyB133DGfeOKJNvsuXLgwTzvttGxsbMzGxsY87bTTcuHChZmZ+eCDD2ZTU1Neeuml2bdv39xqq63yxhtvbD3OH/7wh9xhhx2yoaEh+/fvnz//+c9Xen6TJk3Knj17Zrdu3bK+vj579eqVmZnz58/PESNGZJ8+fXLgwIF5wQUX5JIlSzIz86abbsq999679RhnnHFG7r333jl//vycP39+HnfccbnVVltl//798+yzz87Fixcvt9/pp5+evXv3zsGDB+e4ceNaj3PTTTfl1ltvnQ0NDTl48OC8/fbb2/yZjBo1Ko866qjW9UceeST33HPP7NWrV+6888754IMPtm774he/mOecc07utdde2dDQkF/5yldy7ty5rdtvueWWHDhwYG6++eZ5/vnnV/xzOvroozMicsMNN8z6+vq85JJL2qz13/7t33KrrbbKxsbGvOGGGxLIyZMnZ2bmyJEj8+yzz87MzLlz5+ZBBx2UvXr1ys022yz32WefXLJkSZvneeWVVxLI66+/Ppubm/Pzn/98a9uiRYtan/dZZ52Vu+++e26yySb5ta99Ld94443lXjdly573vffem3V1ddmjR4+sr6/PnXfeufV4v/rVrzIzc8mSJXnBBRfkwIEDs2/fvjlixIicP39+ZmZrHTfffHM2NzfnFltskRdeeGGb3xtJ67Q2M+Y6PYI8b968WpcgrTMmTJjAwoUL+frXv75ce0NDA8OGDWP8+PGtbWPHjuWwww7jzTff5Jvf/CaHHHIIixYtavO4d999N0cccQTz58/na1/7Gt/97nfb7PfTn/6URx99lKeeeoqnn36axx9/nAsvvLB1++uvv87bb7/NjBkzuOGGGzj55JN56623ADj++OO59tpreffdd3nuuecYOnToSsffYYcduOaaa9hzzz1ZsGAB8+fPB+CUU07h7bff5uWXX+ahhx7i1ltv5aabblpu36VLl3LCCSfwzDPP8Mc//pFevXpxzDHH0KNHD6ZMmcJf//pX/vjHP3L99de37vPYY4/xiU98gnnz5nHmmWdy/PHHk5m89957nHrqqdx77728++67TJgwgV133XV1PxoAZsyYwUEHHcQ555zDm2++yaWXXso3vvEN5s6d29rn3//937npppuYM2cOH374IZdeeikAkyZN4jvf+Q6jR49m1qxZrd/HSn5Ot912GwMHDuSee+5hwYIFnHnmmSvVdt9993HppZcyfvx4Jk+ezJ/+9KdVPo/LLruMAQMGMHfuXGbPns1FF11ERKz2PA899BAvvPAC999/f5vHvPXWW7nxxhuZNWsWPXr04NRTT233+3nAAQfw4x//mMMPP5wFCxbw9NNPr9Tn5ptv5uabb+bBBx/k5ZdfZsGCBSu9fh9++GFefPFFHnjgAc4//3xeeOGFds8tad23TgdkSZWbN28effr0oUePlf+L3djYuNwfnP/4j//IoYceSl1dHT/4wQ9YuHAhjz76aJvH3WeffRg2bBjdu3dnxIgRbQYRgNGjR3Puueey5ZZb0rdvX0aNGsVtt93Wur2uro5zzz2Xuro6hg0bRkNDAy+++GLrtkmTJvHOO++w2Wabsdtuu1X0nJcsWcIdd9zBz372MzbZZBMGDx7M6aefvtx5Fy1axJFHHsmbb77JPffcw8Ybb8zs2bMZN24cV155JfX19Wy55ZZ8//vf54477mjdb9CgQZxwwgl0796dkSNHMmvWLGbPng1At27deO655/j73/9OY2MjO+20U7u13n777QwbNoxhw4bRrVs3vvKVrzBkyBDGjRvX2ufYY49l++23Z6ONNmL48OE89dRTAPz2t7/ln//5n9lnn33YYIMNOP/884mI5Y5f6c+pLXfeeSfHHnssn/rUp6ivr+e8885bZd+6ujpmzZrF1KlTqaur4/Of//xKtazovPPOo76+no022qjN7SNGjGg99wUXXMCdd97JkiUfe4oyo0eP5gc/+AHbbLMNDQ0N/OxnP+OOO+5g8eLFrX1GjRrFRhttxC677MIuu+yyRt83SesuA7LURfTp04d58+Yt98t/mVmzZtGnT5/W9ebm5tblbt26MWDAAGbOnNnmcbfaaqvW5Y033piFCxe2eY6ZM2cyaNCg1vVBgwYtd8wttthiufC+8cYbs2DBAgB+97vfMW7cOAYNGsQXv/hFHnnkkUqeMvPmzWPRokUrnbc8ujplyhTGjh3LqFGj2GCDlmnYU6dOZdGiRTQ2NtK7d2969+7NSSedxJw5c1b5vAEWLFhAfX09Y8aM4ZprrqGxsZGDDjqIv/3tb+3WOnXqVH7zm9+0nq937948/PDDzJo1a5XnXPb9mTlz5nI/s4033pgttthiueNX+nNqy4rHL38/V/TDH/6Qbbfdlv33359tttmGiy++uN3jl4/d3vZBgwaxaNGitfIfxLZek4sXL279QwdW/T2XtH4zIEtdxJ577knPnj256667lmtfsGAB9957L/vuu29r27Rp01qXly5dyvTp0+nfv//HOn///v2ZOnVq6/prr71W8TF33313xo4dy5w5czjkkEMYPnx4m/1WHKns06cPdXV1K523qampdX2HHXbgpptu4sADD2wdsW5ubqZnz57MmzeP+fPnM3/+fN555x2ef/75iur96le/yvjx45k1axaf/OQnOeGEE9rdp7m5mREjRrSeb/78+bz33nucddZZ7e7b2NjI9OnTW9f//ve/88Ybb1RUK6z8fWvr+OXXxGuvvbbKvptssgmXXXYZL7/8MnfffTeXX345DzzwwGrP0975Vzx3XV0dffr0ob6+nvfff79125IlS5abktLecdt6Tfbo0YN+/fqtdj9J6z8DstRF9OrVi1GjRnHKKadw3333sWjRIl599VWGDx/OgAEDGDFiRGvfJ598krvuuovFixdz5ZVX0rNnT/bYY4+Pdf4jjzySCy+8kLlz5zJv3jzOP/98jj766Hb3+/DDDxk9ejRvv/02dXV1bLrppnTr1vZbV79+/Zg+fToffvghAN27d2f48OGcffbZvPvuu0ydOpXLL798pfMeeeSRXHTRRey333689NJLNDY2sv/++3P66afzzjvvsHTpUl566SUeeuihduudPXs2Y8eO5b333qNnz540NDSsst6yo48+mnvuuYf777+fJUuWsHDhQv785z8vF3xX5dBDD+Wee+5hwoQJfPjhh5x33nlkZrv7LdOvXz9efvnlVW4fPnw4N998M5MmTeL999/nJz/5ySr7/v73v2fKlClkJr169aJ79+6tz7+986zK7bff3nruc889l0MPPZTu3buz/fbbs3DhQv7whz+waNEiLrzwQj744IPlnterr77K0qVL2zzukUceyRVXXMErr7zCggULWucstzUNSVLXYkCWupAzzzyTiy66iDPOOINNN92Uz33uczQ3N/PAAw/Qs2fP1n4HH3wwY8aMYbPNNuO2227jrrvuoq6u7mOd+5xzzmHIkCHsvPPOfPrTn2a33XbjnHPOqWjf2267jcGDB7PppptyzTXXMHr06Db7DR06lJ122omtttqqdcrIVVddRX19Pdtssw377LMP3/zmNznuuONW2nfkyJGce+65DB06lFdffZVbb72VDz/8kB133JHNNtuMQw89dLnpDquydOlSLr/8cvr378/mm2/OQw89xNVXX93ufs3NzYwdO5aLLrqIvn370tzczM9//vNVhruynXbaiauuuoojjjiCxsZGGhoa2HLLLZf7ma7Oj370Iy688EJ69+7d+sG/sgMPPJDvfe97DB06lG233bbND0kuM3nyZPbbbz8aGhrYc889+c53vsOXv/zlis6zKiNGjOCYY45hq622YuHChfziF78AWv7o++Uvf8m3v/1tmpqaqK+vZ8CAAa37HXbYYUDL9J225q0fd9xxjBgxgi984QtsvfXWbLjhhlx11VUV1yVp/RVrMsrQ2QwZMiQnTpxY6zKk9cp5553HlClTuP32qt/gTVWyYMECevfuzeTJk9l6661rXY4kdWZtzsVyBFmS1gP33HMP77//Pu+99x5nnHEGn/70pxk8eHCty5KkdZIBWZLWA2PHjqV///7079+fyZMnc8cdd7T7ITVJUtucYiFJkqSuyikWkiRJUnsMyJIkSVKJAVmSJEkqMSBLkiRJJQZkfSxNzQOJiIoeTc0Da12uJElSu7yfpj6WmdOncfi1EyrqO+akvapcjSRJ0sfnCLIkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSoxIEuSJEklBmRJkiSpxIAsSZIklRiQJUmSpBIDsiRJklRiQJYkSZJKDMiSJElSiQFZkiRJKjEgS5IkSSVVC8gRsWFEPB4RT0fE8xHxk6J964h4LCKmRMSYiNigaO9ZrE8ptg+uVm2SJEnSqlRzBPkDYGhm7gLsChwQEXsAlwBXZOa2wFvA8UX/44G3ivYrin6SJElSh6paQM4WC4rVuuKRwFDgt0X7LcAhxfLBxTrF9n0jIqpVnyRJktSWqs5BjojuEfEUMAcYD7wEzM/MxUWX6UBTsdwETAMotr8NbNHGMU+MiIkRMXHu3LnVLF+SJEldUFUDcmYuycxdgQHAZ4FProVjXpeZQzJzSN++fT92jZIkSVJZh1zFIjPnAw8CewK9I6JHsWkAMKNYngE0AxTbewFvdER9kiRJ0jLVvIpF34joXSxvBHwFeIGWoHxo0W0kMLZYvrtYp9j+n5mZ1apPkiRJakuP9rt8ZI3ALRHRnZYgfmdm/j4iJgF3RMSFwF+BG4r+NwC3RcQU4E3giCrWJkmSJLWpagE5M58BPtNG+8u0zEdesX0hcFi16pEkSZIq4Z30JEmSpBIDsiRJklRiQJYkSZJKDMjqON16EBEVPZqaB9a6WkmS1EVV8yoW0vKWLubwaydU1HXMSXtVuRhJkqS2OYIsSZIklRiQu4Cm5oEVT21weoMkSerqnGLRBcycPq3iqQ3g9AZJktS1OYIsSZIklRiQJUmSpBIDsiRJklRiQJYkSZJKDMiSJElSiQFZkiRJKjEgS5IkSSUGZEmSJKnEgCxJkiSVGJAlSZKkEgOyJEmSVGJAXkc1NQ8kIip6SJIkqXI9al2APpqZ06dx+LUTKuo75qS9qlyNJEnS+sMRZEmSJKnEgCxJkiSVGJAlSZKkEgOyJEmSVGJAliRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSqpWkCOiOaIeDAiJkXE8xFxWtF+XkTMiIinisew0j4/iogpEfFiRHy1WrVJkiRJq9KjisdeDJyemX+JiE2AJyNifLHtisy8tNw5InYEjgB2AvoDf4qI7TNzSRVrlCRJkpZTtRHkzJyVmX8plt8FXgCaVrPLwcAdmflBZr4CTAE+W636JEmSpLZ0yBzkiBgMfAZ4rGj6bkQ8ExE3RsRmRVsTMK2023RWH6glSZKkta7qATkiGoDfAd/LzHeAq4F/AHYFZgGXreHxToyIiRExce7cuWu9XkmSJHVtVQ3IEVFHSzgenZl3AWTm7MxckplLgV/x39MoZgDNpd0HFG3LyczrMnNIZg7p27dvNcuXJElSF1TNq1gEcAPwQmZeXmpvLHX7F+C5Yvlu4IiI6BkRWwPbAY9Xqz5JkiSpLdW8isXewAjg2Yh4qmj7MXBkROwKJPAqcBJAZj4fEXcCk2i5AsbJXsFClWhqHsjM6dPa7wj0H9DMjGmvVbkiSZK0LqtaQM7Mh4FoY9O41ezzU+Cn1apJ66eZ06dx+LUTKuo75qS9qlyNJEla13knPUmSJKnEgCxJkiSVGJAlSZKkEgOyJEmSVGJAliRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSoxIEuSJEklBmRJkiSpxIAsSZIklRiQJUmSpBIDsiRJklRiQJYkSZJKDMiSJElSiQFZkiRJKjEgS5IkSSUGZEmSJKnEgCxJkiSVGJAlSZKkEgOyJEmSVGJAliRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSqoWkCOiOSIejIhJEfF8RJxWtG8eEeMjYnLxdbOiPSLiFxExJSKeiYjdqlWbJEmStCrVHEFeDJyemTsCewAnR8SOwFnAA5m5HfBAsQ5wILBd8TgRuLqKtUmSJEltqlpAzsxZmfmXYvld4AWgCTgYuKXodgtwSLF8MHBrtngU6B0RjdWqT5IkSWpLh8xBjojBwGeAx4B+mTmr2PQ60K9YbgKmlXabXrSteKwTI2JiREycO3du1WqWJElS11T1gBwRDcDvgO9l5jvlbZmZQK7J8TLzuswckplD+vbtuxYrlSRJkqockCOijpZwPDoz7yqaZy+bOlF8nVO0zwCaS7sPKNokSZKkDlPNq1gEcAPwQmZeXtp0NzCyWB4JjC21f6u4msUewNulqRiSJElSh+hRxWPvDYwAno2Ip4q2HwMXA3dGxPHAVGB4sW0cMAyYArwPHFvF2tRVdetBy99u7es/oJkZ016rckGSJKmzqVpAzsyHgVUlkX3b6J/AydWqRwJg6WIOv3ZCRV3HnLRXlYuRJEmdkXfSkyRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSqpKCBHxN6VtEmSJEnrukpHkK+qsE2SJElap/VY3caI2BPYC+gbET8obdoU6F7NwiRJkqRaWG1ABjYAGop+m5Ta3wEOrVZRkiRJUq2sNiBn5kPAQxFxc2ZO7aCaJEmSpJppbwR5mZ4RcR0wuLxPZg6tRlGSJElSrVQakH8DXANcDyypXjmSJElSbVUakBdn5tVVrUSSJEnqBCq9zNs9EfGdiGiMiM2XPapamSRJklQDlY4gjyy+/rDUlsA2a7ccSZIkqbYqCsiZuXW1C5EkSZI6g4oCckR8q632zLx17ZYjSZIk1ValUyx2Ly1vCOwL/AUwIEuSJGm9UukUi1PK6xHRG7ijKhVJ66Cm5oHMnD6tor79BzQzY9prVa5IkiR9VJWOIK/oPcB5yVJh5vRpHH7thIr6jjlprypXI0mSPo5K5yDfQ8tVKwC6AzsAd1arKEmSJKlWKh1BvrS0vBiYmpnTq1CPJEmSVFMV3SgkMx8C/gZsAmwGfFjNoiRJkqRaqSggR8Rw4HHgMGA48FhEHFrNwiRJkqRaqHSKxdnA7pk5ByAi+gJ/An5brcIkSZKkWqhoBBnotiwcF95Yg30lSZKkdUalI8j3RcT9wK+L9cOBcdUpSZIkSaqd1QbkiNgW6JeZP4yIrwP7FJseAUZXuzhJkiSpo7U3gnwl8COAzLwLuAsgIj5dbPvnqlYnSZIkdbD25hH3y8xnV2ws2gZXpSJJkiSphtoLyL1Xs22j1e0YETdGxJyIeK7Udl5EzIiIp4rHsNK2H0XElIh4MSK+Wln5kiRJ0trVXkCeGBEnrNgYEd8Gnmxn35uBA9povyIzdy0e44rj7QgcAexU7PPLiOjeXvGSJEnS2tbeHOTvAf8REUfx34F4CLAB8C+r2zEz/3dEDK6wjoOBOzLzA+CViJgCfJaWDwNKkiRJHWa1ATkzZwN7RcSXgU8VzX/IzP/8GOf8bkR8C5gInJ6ZbwFNwKOlPtOLNkmSJKlDVXSzj8x8MDOvKh4fJxxfDfwDsCswC7hsTQ8QESdGxMSImDh37tyPUYokSZK0sg69G15mzs7MJZm5FPgVLdMoAGYAzaWuA4q2to5xXWYOycwhffv2rW7BkiRJ6nI6NCBHRGNp9V+AZVe4uBs4IiJ6RsTWwHbA4x1ZmyRJkgSV32p6jUXEr4EvAX0iYjowCvhSROwKJPAqcBJAZj4fEXcCk4DFwMmZuaRatUmSJEmrUrWAnJlHttF8w2r6/xT4abXqkSRJkipRtYCsdVi3HkRErauQJEmqCQOyVrZ0MYdfO6GirmNO2qvKxUiSJHWsDv2QniRJktTZGZAlSZKkEgOyJEmSVGJAliRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJV4HWZ2TNyuRJEk1YkBW5+TNSiRJUo04xUKSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkDuRJqaBxIRFT0kSZJUHV4HuROZOX2a1/6VJEmqMUeQJUmSpBIDsiRJklRiQJY6WrceFc81b2oeWOtqJUnqcpyDLHW0pYuday5JUifmCLIkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSoxIEuSJEklBmRJkiSpxIAsSZIklRiQJUmSpBIDsiRJklRiQJYkSZJKDMiSJElSiQFZWk80NQ8kIip6NDUPrHW5kiR1Wj1qXYCktWPm9Gkcfu2EivqOOWmvKlcjSdK6yxFkSZIkqaRqATkiboyIORHxXKlt84gYHxGTi6+bFe0REb+IiCkR8UxE7FatuiRJkqTVqeYI8s3AASu0nQU8kJnbAQ8U6wAHAtsVjxOBq6tYlyRJkrRKVQvImfm/gTdXaD4YuKVYvgU4pNR+a7Z4FOgdEY3Vqk2SJElalY6eg9wvM2cVy68D/YrlJmBaqd/0ok2SJEnqUDX7kF5mJpBrul9EnBgREyNi4ty5c6tQ2dq1JpfekiRJUu119GXeZkdEY2bOKqZQzCnaZwDNpX4DiraVZOZ1wHUAQ4YMWeOA3dG89JYkSdK6paNHkO8GRhbLI4GxpfZvFVez2AN4uzQVQ5IkSeowVRtBjohfA18C+kTEdGAUcDFwZ0QcD0wFhhfdxwHDgCnA+8Cx1apLkiRJWp2qBeTMPHIVm/Zto28CJ1erFkmSJKlS3klPkiRJKjEgS5IkSSUGZEmSJKnEgCxJkiSVGJAlSZKkEgOyJEmSVGJAliRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIUlfUrQcRUfGjqXlgrSuWJKnD9Kh1AZJqYOliDr92QsXdx5y0VxWLkSSpc3EEWZIkSSoxIEuSJEklBmRJkiSpxIAsSZIklfghPWlViis9SJKkrsWALK3KGlzpwas8SJK0/nCKhSRJklRiQJYkSZJKDMiSJElSiQFZkiRJKvFDelJn5pU0JEnqcAZkqTPzShqSJHU4p1hIkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSqpyY1CIuJV4F1gCbA4M4dExObAGGAw8CowPDPfqkV9kiRJ6rpqOYL85czcNTOHFOtnAQ9k5nbAA8W6JEmS1KE60xSLg4FbiuVbgENqWIskSZK6qFoF5AT+GBFPRsSJRVu/zJxVLL8O9Gtrx4g4MSImRsTEuXPndkStkiRJ6kJqMgcZ2CczZ0TElsD4iPhbeWNmZkRkWztm5nXAdQBDhgxps48kSZL0UdVkBDkzZxRf5wD/AXwWmB0RjQDF1zm1qE2SJEldW4cH5Iioj4hNli0D+wPPAXcDI4tuI4GxHV2bJEmSVIspFv2A/4iIZef/98y8LyKeAO6MiOOBqcDwGtQmSZKkLq7DA3Jmvgzs0kb7G8C+HV2PpLWrqXkgM6dPq6hv/wHNzJj2WpUrkiRpzdTqQ3qS1lMzp0/j8GsnVNR3zEl7VbkaSZLWXGe6DrIkSZJUc44gS2pftx4UnxuQJGm9Z0CW1L6li502IUnqMpxiIUmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSoxIEta7zQ1DyQiKno0NQ+sdbmSpE7G6yBLWu94u2tJ0sfhCLIkSZJUYkCWJEmSSgzIkmqnWw/nCkuSOh3nIEuqnaWLK58r/D++QERUuSBJkgzIktYVaxKm/eCdJOljcIqFJEmSVGJAliRJkkoMyJIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVJkqQSA7IkSZJUYkCWJEmSSgzIkiRJUokBWZIkSSoxIH8ETc0DiYiKHpIkSVq39Kh1AeuimdOncfi1EyrqO+akvapcjaR1XVPzQGZOn1ZR3/4Dmpkx7bUqVyRJXZsBWZIqtCZBtntdT5Ys+qDiY69rf3Qb6iWtzwzIklShNf3vUVVCb7ceFU/fqmYw9T9pktZnBmRJXdsaBM5OYenidS+YdpJQL0mVMiBL6trWxcBZBWsyZWKN+T2WuqR1eSqWAVmS1shMxP8AAAg7SURBVFdrODpeaYgFg6yk9q3LU7E6XUCOiAOA/wl0B67PzItrXJIkrZscua26dXmETNKqdaqAHBHdgf8FfAWYDjwREXdn5qTaViZJ0sqqNUJm8JZqq1MFZOCzwJTMfBkgIu4ADgYMyJLUFXSCD/RVdT52hdblf01L64POFpCbgPK70nTgczWqRZLU0TrBtBDDqaTIzFrX0CoiDgUOyMxvF+sjgM9l5ndLfU4ETixWPwG8WKVy+gDzqnRsrdt8bagtvi60Kr421BZfF53DvMw8YMXGzjaCPANoLq0PKNpaZeZ1wHXVLiQiJmbmkGqfR+seXxtqi68LrYqvDbXF10Xn1q3WBazgCWC7iNg6IjYAjgDurnFNkiRJ6kI61QhyZi6OiO8C99NymbcbM/P5GpclSZKkLqRTBWSAzBwHjKt1HXTANA6ts3xtqC2+LrQqvjbUFl8XnVin+pCeJEmSVGudbQ6yJEmSVFMG5DZExAER8WJETImIs2pdj2ojIpoj4sGImBQRz0fEaUX75hExPiImF183q3Wtqo2I6B4Rf42I3xfrW0fEY8V7x5jiw8bqQiKid0T8NiL+FhEvRMSevmcIICK+X/wueS4ifh0RG/qe0XkZkFdQut31gcCOwJERsWNtq1KNLAZOz8wdgT2Ak4vXwlnAA5m5HfBAsa6u6TTghdL6JcAVmbkt8BZwfE2qUi39T+C+zPwksAstrw/fM7q4iGgCTgWGZOanaLkQwRH4ntFpGZBX1nq768z8EFh2u2t1MZk5KzP/Uiy/S8svuiZaXg+3FN1uAQ6pTYWqpYgYABwEXF+sBzAU+G3RxddGFxMRvYAvADcAZOaHmTkf3zPUogewUUT0ADYGZuF7RqdlQF5ZW7e7bqpRLeokImIw8BngMaBfZs4qNr0O9KtRWaqtK4EzgaXF+hbA/MxcXKz73tH1bA3MBW4qpt5cHxH1+J7R5WXmDOBS4DVagvHbwJP4ntFpGZCldkREA/A74HuZ+U55W7ZcBsZLwXQxEfFPwJzMfLLWtahT6QHsBlydmZ8B3mOF6RS+Z3RNxbzzg2n5I6o/UA+sdHtjdR4G5JW1e7trdR0RUUdLOB6dmXcVzbMjorHY3gjMqVV9qpm9ga9FxKu0TMMaSsvc097Fv0/B946uaDowPTMfK9Z/S0tg9j1D+wGvZObczFwE3EXL+4jvGZ2UAXll3u5aQOuc0huAFzLz8tKmu4GRxfJIYGxH16bayswfZeaAzBxMy3vEf2bmUcCDwKFFN18bXUxmvg5Mi4hPFE37ApPwPUMtUyv2iIiNi98ty14bvmd0Ut4opA0RMYyW+YXLbnf90xqXpBqIiH2A/wKe5b/nmf6YlnnIdwIDganA8Mx8syZFquYi4kvAGZn5TxGxDS0jypsDfwWOzswPalmfOlZE7ErLBzc3AF4GjqVlMMr3jC4uIn4CHE7LFZL+CnybljnHvmd0QgZkSZIkqcQpFpIkSVKJAVmSJEkqMSBLkiRJJQZkSZIkqcSALEmSJJUYkCVpDURERsRlpfUzIuK8tXTsmyPi0PZ7fuzzHBYRL0TEgyu0D46Ib1aw/zER8f9Vr0JJqi0DsiStmQ+Ar0dEn1oXUla6G1cljgdOyMwvr9A+GGg3IEvS+s6ALElrZjFwHfD9FTesOAIcEQuKr1+KiIciYmxEvBwRF0fEURHxeEQ8GxH/UDrMfhExMSL+/4j4p2L/7hHx84h4IiKeiYiTSsf9r4i4m5a7cq1Yz5HF8Z+LiEuKtnOBfYAbIuLnK+xyMfD5iHgqIr4fERtGxE3FMf4aESsGaiLioIh4JCL6RMT+xfJfIuI3EdFQ9Hk1In5StD8bEZ8s2r9YnOup4vibVP5jkKTqMSBL0pr7X8BREdFrDfbZBfhXYAdgBLB9Zn6WlruunVLqNxj4LHAQcE1EbEjLiO/bmbk7sDtwQkRsXfTfDTgtM7cvnywi+gOXAEOBXYHdI+KQzDwfmAgclZk/XKHGs4D/ysxdM/MK4GQgM/PTwJHALUU9y87xL8U+w4qmc4D9MnO34hw/KB17XtF+NXBG0XYGcHJm7gp8Hvh7e99ESeoIBmRJWkOZ+Q5wK3DqGuz2RGbOKm4j+xLwx6L9WVpC8TJ3ZubSzJxMy62KPwnsD3wrIp6i5VbnWwDbFf0fz8xX2jjf7sCfM3NuZi4GRgNfWIN6oWWk+XaAzPwbLbdJXhbEhwL/D3BQZr4F7AHsCPyfos6RwKDSse4qvj5Zer7/B7g8Ik4Fehd1SlLNGZAl6aO5kpaR3fpS22KK99WI6AZsUNr2QWl5aWl9KVCeP5wrnCeBAE4pRnZ3zcytM3NZwH7vYz2Lj+4lYBP+OzAHML5U446ZeXyp/7Lnu4Ti+WbmxcC3gY1oCdaf7JjSJWn1DMiS9BFk5pvAnbSE5GVeBf6xWP4aUPcRDn1YRHQr5iVvA7wI3A/8j4ioA4iI7SOifnUHAR4HvljMDe5OyxSJh9rZ511aQu8y/wUcteycwMCiHmgZTf4GcGtE7AQ8CuwdEdsW/euLfVYpIv4hM5/NzEuAJ2gZLZekmjMgS9JHdxlQvprFr2gJpU8De/LRRndfoyXc3gv8a2YupGWe8iTgLxHxHHAty486ryQzZ9EyP/hB4Gngycwc2865nwGWRMTTEfF94JdAt4h4FhgDHFNMEVl2jr/REqB/A2wKHAP8OiKeAR6h/cD7veIDhM8Ai4rnLEk1F5kr/jdPkiRJ6rocQZYkSZJKDMiSJElSiQFZkiRJKjEgS5IkSSUGZEmSJKnEgCxJkiSVGJAlSZKkEgOyJEmSVPJ/AW4aZ+X17Oo+AAAAAElFTkSuQmCC\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from matplotlib import pyplot as plt\n",
        "fig, axes = plt.subplots(1,2,figsize=(10,5))\n",
        "df['tag_len'].plot.hist(ax=axes[0])\n",
        "df.boxplot(column=['tag_len'],ax=axes[1])\n",
        "axes[0].set_title(\"Sentences tokens count distribution\")\n",
        "axes[1].set_title(\"Boxplot Sentences tokens\")\n",
        "axes[0].set_xlabel('number of tokens')\n",
        "axes[0].set_ylabel('frequency')\n",
        "axes[1].set_ylabel('number of tokens')\n",
        "plt.savefig(\"op_tokens_len.png\")\n",
        "print(\"min:{} max:{} avg:{}\".format(df['tag_len'].min(),df['tag_len'].max(),df['tag_len'].mean()))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 367
        },
        "id": "X7nRhA3Z7OeI",
        "outputId": "33f3728c-39bc-4a25-bb2d-56907e975f0f"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "min:1 max:90 avg:16.232712765957448\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "fig, ax = plt.subplots(figsize=(7,7))\n",
        "df['pol'].value_counts().plot.pie(ax=ax, autopct=\"%.1f%%\");\n",
        "plt.title(\"Polarity distribution of the dataset.\")\n",
        "plt.savefig(\"polarity_distribution.png\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 427
        },
        "id": "4-XsBTM250qT",
        "outputId": "b9099671-161e-44c9-c613-507519ecbde8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 504x504 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Just Aspects"
      ],
      "metadata": {
        "id": "UMGHVW79BznX"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install -q tf-models-official"
      ],
      "metadata": {
        "id": "2eiQaBV8Xdu_",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "da276391-1570-4157-abdc-31f7c11b98b9"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[K     |████████████████████████████████| 2.1 MB 4.4 MB/s \n",
            "\u001b[K     |████████████████████████████████| 43 kB 1.4 MB/s \n",
            "\u001b[K     |████████████████████████████████| 237 kB 62.7 MB/s \n",
            "\u001b[K     |████████████████████████████████| 1.1 MB 50.7 MB/s \n",
            "\u001b[K     |████████████████████████████████| 636 kB 51.9 MB/s \n",
            "\u001b[K     |████████████████████████████████| 1.2 MB 42.8 MB/s \n",
            "\u001b[K     |████████████████████████████████| 92 kB 7.3 MB/s \n",
            "\u001b[K     |████████████████████████████████| 352 kB 58.5 MB/s \n",
            "\u001b[K     |████████████████████████████████| 47.8 MB 2.3 MB/s \n",
            "\u001b[K     |████████████████████████████████| 99 kB 8.8 MB/s \n",
            "\u001b[?25h  Building wheel for py-cpuinfo (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Building wheel for seqeval (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "import re\n",
        "import json\n",
        "import string\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "from sklearn import preprocessing\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "# from official.nlp import optimization\n",
        "\n",
        "max_len=266\n",
        "query_len=10\n",
        "max_len_txt=max_len-query_len"
      ],
      "metadata": {
        "id": "PNjD8bj6By-e"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "df = pd.read_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")"
      ],
      "metadata": {
        "id": "kQUFlo5ltgJb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#For loading hub layers from HUB in TPUs\n",
        "os.environ[\"TFHUB_MODEL_LOAD_FORMAT\"] = \"UNCOMPRESSED\""
      ],
      "metadata": {
        "id": "j_IaKwTPwb6c"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "epochs=3\n",
        "def create_model_class():\n",
        "    ## BERT encoder\n",
        "    #encoder\n",
        "    encoder = TFBertModel.from_pretrained(\"bert-base-uncased\")\n",
        "    ## NER Model\n",
        "    input_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    token_type_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    attention_mask = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    embedding = encoder(\n",
        "        input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask\n",
        "    )['last_hidden_state'][:,0]\n",
        "    # ['pooler_output']\n",
        "    embedding = layers.Dropout(0.1)(embedding)\n",
        "    l = layers.Dense(32, activation=\"tanh\")(embedding)\n",
        "    # output = layers.Dense(1, activation='sigmoid')(l)\n",
        "    output = layers.Dense(2, activation='softmax')(l)\n",
        "    \n",
        "    model = keras.Model(\n",
        "        inputs=[input_ids, token_type_ids, attention_mask],\n",
        "        outputs=[output],\n",
        "    )\n",
        "    METRICS = [\n",
        "              tf.keras.metrics.CategoricalAccuracy(),\n",
        "              tf.keras.metrics.Precision(),\n",
        "              tf.keras.metrics.Recall()\n",
        "    ]\n",
        "    loss = tf.keras.losses.BinaryCrossentropy(from_logits=True)\n",
        "    metrics = tf.metrics.BinaryAccuracy()    \n",
        "    steps_per_epoch = df.shape[0]\n",
        "    num_train_steps = steps_per_epoch * epochs\n",
        "    num_warmup_steps = int(0.1*num_train_steps)\n",
        "\n",
        "    init_lr = 3e-5\n",
        "    \"\"\"optimizer = optimization.create_optimizer(init_lr=init_lr,\n",
        "                                              num_train_steps=num_train_steps,\n",
        "                                              num_warmup_steps=num_warmup_steps,\n",
        "                                              optimizer_type='adamw')\"\"\"\n",
        "    # model.compile(loss=loss, optimizer=optimizer, metrics=metrics)\n",
        "    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0), \n",
        "              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), \n",
        "              metrics=[tf.keras.metrics.SparseCategoricalAccuracy('accuracy')])\n",
        "\n",
        "    return model"
      ],
      "metadata": {
        "id": "YWkV2IsVD1Dd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "use_tpu = None\n",
        "try:\n",
        "    tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n",
        "    use_tpu = True\n",
        "except:\n",
        "    use_tpu = False\n",
        "\n",
        "if use_tpu:\n",
        "    # Create distribution strategy\n",
        "    tf.config.experimental_connect_to_cluster(tpu)\n",
        "    tf.tpu.experimental.initialize_tpu_system(tpu)\n",
        "    strategy = tf.distribute.experimental.TPUStrategy(tpu)\n",
        "\n",
        "    # Create model\n",
        "    with strategy.scope():\n",
        "        model = create_model_class()\n",
        "else:\n",
        "    model = create_model_class()\n",
        "    \n",
        "model.summary()"
      ],
      "metadata": {
        "id": "GZwWa7aiFB-b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9479225f-a475-4591-df43-35a958b8423d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Deallocate tpu buffers before initializing tpu system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Deallocate tpu buffers before initializing tpu system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:TPU system grpc://10.12.21.138:8470 has already been initialized. Reinitializing the TPU can cause previously created variables on TPU to be lost.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:tensorflow:TPU system grpc://10.12.21.138:8470 has already been initialized. Reinitializing the TPU can cause previously created variables on TPU to be lost.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Initializing the TPU system: grpc://10.12.21.138:8470\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Initializing the TPU system: grpc://10.12.21.138:8470\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Finished initializing TPU system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Finished initializing TPU system.\n",
            "WARNING:absl:`tf.distribute.experimental.TPUStrategy` is deprecated, please use  the non experimental symbol `tf.distribute.TPUStrategy` instead.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Found TPU system:\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Found TPU system:\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Num TPU Workers: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Num TPU Workers: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores Per Worker: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores Per Worker: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:localhost/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:localhost/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:0, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:0, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:1, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:1, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:2, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:2, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:3, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:3, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:4, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:4, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:5, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:5, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:6, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:6, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:7, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:7, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU_SYSTEM:0, TPU_SYSTEM, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU_SYSTEM:0, TPU_SYSTEM, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:XLA_CPU:0, XLA_CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:XLA_CPU:0, XLA_CPU, 0, 0)\n",
            "Some layers from the model checkpoint at bert-base-uncased were not used when initializing TFBertModel: ['mlm___cls', 'nsp___cls']\n",
            "- This IS expected if you are initializing TFBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing TFBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "All the layers of TFBertModel were initialized from the model checkpoint at bert-base-uncased.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFBertModel for predictions without further training.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model_2\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_7 (InputLayer)           [(None, 266)]        0           []                               \n",
            "                                                                                                  \n",
            " input_9 (InputLayer)           [(None, 266)]        0           []                               \n",
            "                                                                                                  \n",
            " input_8 (InputLayer)           [(None, 266)]        0           []                               \n",
            "                                                                                                  \n",
            " tf_bert_model_2 (TFBertModel)  TFBaseModelOutputWi  109482240   ['input_7[0][0]',                \n",
            "                                thPoolingAndCrossAt               'input_9[0][0]',                \n",
            "                                tentions(last_hidde               'input_8[0][0]']                \n",
            "                                n_state=(None, 266,                                               \n",
            "                                 768),                                                            \n",
            "                                 pooler_output=(Non                                               \n",
            "                                e, 768),                                                          \n",
            "                                 past_key_values=No                                               \n",
            "                                ne, hidden_states=N                                               \n",
            "                                one, attentions=Non                                               \n",
            "                                e, cross_attentions                                               \n",
            "                                =None)                                                            \n",
            "                                                                                                  \n",
            " tf.__operators__.getitem_2 (Sl  (None, 768)         0           ['tf_bert_model_2[0][0]']        \n",
            " icingOpLambda)                                                                                   \n",
            "                                                                                                  \n",
            " dropout_113 (Dropout)          (None, 768)          0           ['tf.__operators__.getitem_2[0][0\n",
            "                                                                 ]']                              \n",
            "                                                                                                  \n",
            " dense_4 (Dense)                (None, 32)           24608       ['dropout_113[0][0]']            \n",
            "                                                                                                  \n",
            " dense_5 (Dense)                (None, 2)            66          ['dense_4[0][0]']                \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 109,506,914\n",
            "Trainable params: 109,506,914\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def tokenize_couple(ex):\n",
        "  dct=slow_tokenizer(ex[0], max_length=max_len_txt, padding='max_length')\n",
        "  dct2=slow_tokenizer(ex[1]+\"[SEP]\",add_special_tokens=False,max_length=query_len, padding='max_length')\n",
        "  dct2['token_type_ids']=[1 for i in range(query_len)]\n",
        "  dct['input_ids'].extend(dct2['input_ids'])\n",
        "  dct['token_type_ids'].extend(dct2['token_type_ids'])\n",
        "  dct['attention_mask'].extend(dct2['attention_mask'])\n",
        "  return dct"
      ],
      "metadata": {
        "id": "F21WnnHfVt9J"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def create_inputs_targets():\n",
        "  input_ids= []\n",
        "  token_type_ids= []\n",
        "  attention_mask= []\n",
        "  tags = []\n",
        "  f = []\n",
        "  for i, row in df.iterrows():\n",
        "    dic = tokenize_couple((row['txt'],row['categ_txt']))\n",
        "    tagging = [0]+list(map(lambda x: 0 if x=='O' else 1 if x=='B' else 2,row['tagging']))+[0]\n",
        "    tagging.extend([3]*(max_len_txt-len(tagging)-1))\n",
        "    input_ids.append(dic['input_ids'])\n",
        "    token_type_ids.append(dic['token_type_ids'])\n",
        "    attention_mask.append(dic['attention_mask'])\n",
        "    tags.append(tagging)\n",
        "    f.append(1 if row['f']=='yes' else 0)          \n",
        "  x = [\n",
        "      tf.convert_to_tensor(np.array(input_ids), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(token_type_ids), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(attention_mask), dtype=tf.dtypes.int32),\n",
        "  ]\n",
        "  return x, tf.convert_to_tensor(np.array(tags,dtype=np.dtype('i')),dtype=tf.dtypes.int32), tf.convert_to_tensor(np.array(f,dtype=np.dtype('i')),dtype=tf.dtypes.int32)"
      ],
      "metadata": {
        "id": "0BMfzt3sUDOj"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df['f'].value_counts().sort_values().plot(kind = 'barh')\n",
        "df['f'].value_counts()"
      ],
      "metadata": {
        "id": "66wWH6IGEwsl",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 317
        },
        "outputId": "226711e8-77f3-435f-9f41-a6f49d804be8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "no     1504\n",
              "yes    1504\n",
              "Name: f, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 71
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# x_train, y_train, val_x, val_y = create_ds()\n",
        "x_train, y_train_t, y_train = create_inputs_targets()"
      ],
      "metadata": {
        "id": "eWLCWO9nJddn"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "bs=8\n",
        "epochs = 3\n",
        "\"\"\"history = model.fit(\n",
        "    x_train,\n",
        "    y_train,\n",
        "    epochs=1,\n",
        "    verbose=1,\n",
        "    batch_size=bs,\n",
        "    validation_split=0.1\n",
        ")\"\"\"\n",
        "history = model.fit(x=x_train, y=y_train,\n",
        "                    validation_split=0.1, batch_size=bs,\n",
        "                               epochs=epochs)"
      ],
      "metadata": {
        "id": "8z1LDFw6JjnX",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6eb06ecf-486b-4abc-85b4-294343272b69"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/3\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/tensorflow/python/util/dispatch.py:1082: UserWarning: \"`sparse_categorical_crossentropy` received `from_logits=True`, but the `output` argument was produced by a sigmoid or softmax activation and thus does not represent logits. Was this intended?\"\n",
            "  return dispatch_target(*args, **kwargs)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_2/bert/pooler/dense/kernel:0', 'tf_bert_model_2/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_2/bert/pooler/dense/kernel:0', 'tf_bert_model_2/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_2/bert/pooler/dense/kernel:0', 'tf_bert_model_2/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_2/bert/pooler/dense/kernel:0', 'tf_bert_model_2/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "339/339 [==============================] - 142s 125ms/step - loss: 0.4940 - accuracy: 0.7606 - val_loss: 0.4017 - val_accuracy: 0.8538\n",
            "Epoch 2/3\n",
            "339/339 [==============================] - 31s 93ms/step - loss: 0.2670 - accuracy: 0.8932 - val_loss: 0.4445 - val_accuracy: 0.8272\n",
            "Epoch 3/3\n",
            "339/339 [==============================] - 31s 93ms/step - loss: 0.1470 - accuracy: 0.9512 - val_loss: 0.4240 - val_accuracy: 0.8870\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "history_dict = history.history\n",
        "print(history_dict.keys())\n",
        "\n",
        "acc = history_dict['accuracy']\n",
        "val_acc = history_dict['val_accuracy']\n",
        "loss = history_dict['loss']\n",
        "val_loss = history_dict['val_loss']\n",
        "\n",
        "epochs = range(1, len(acc) + 1)\n",
        "fig = plt.figure(figsize=(10, 6))\n",
        "fig.tight_layout()\n",
        "\n",
        "plt.subplot(2, 1, 1)\n",
        "# r is for \"solid red line\"\n",
        "plt.plot(epochs, loss, 'r', label='Training loss')\n",
        "# b is for \"solid blue line\"\n",
        "plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
        "plt.title('Training and validation loss')\n",
        "# plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "\n",
        "plt.subplot(2, 1, 2)\n",
        "plt.plot(epochs, acc, 'r', label='Training acc')\n",
        "plt.plot(epochs, val_acc, 'b', label='Validation acc')\n",
        "plt.title('Training and validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend(loc='lower right')"
      ],
      "metadata": {
        "id": "Wq77BmmUhlln",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 439
        },
        "outputId": "ae1aefde-45e7-4bc2-e5ca-965c91eca253"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7f0c4b56e050>"
            ]
          },
          "metadata": {},
          "execution_count": 85
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x432 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Inference**"
      ],
      "metadata": {
        "id": "tf5PbY2gZBIZ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#Inference\n",
        "# c=(\"The pizza was very good but the service wasn't.\",\"food quality positive\")\n",
        "# c=(\"The pastas are incredible, the risottos (particularly the sepia) are fantastic and the braised rabbit is amazing.\", \"food quality positive\")\n",
        "c=(\"The spaghetti were very tasty.\",\"food quality positive\")\n",
        "tokens=tokenize_couple(c)\n",
        "x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "out = model.predict(x)\n",
        "out"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "B-I5f__7ZAot",
        "outputId": "93c6084a-5f26-492e-ea18-061c89dc5ea9"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[0.00980832, 0.99019176]], dtype=float32)"
            ]
          },
          "metadata": {},
          "execution_count": 86
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# model.save(\"/content/drive/MyDrive/HLT/class_model_balanced.h5\")"
      ],
      "metadata": {
        "id": "dNex3os9DkV-"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vSlujw-M33Su"
      },
      "source": [
        "# Just Entity recognition"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "import re\n",
        "import json\n",
        "import string\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "from sklearn import preprocessing\n",
        "\n",
        "max_len=266\n",
        "query_len=10\n",
        "max_len_txt=max_len-query_len"
      ],
      "metadata": {
        "id": "T7z57oNh8bp9"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "df = pd.read_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")"
      ],
      "metadata": {
        "id": "6mW75jPUkLYS"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "ex = (\"Not impressed with the food.\",\"food quality negative\")\n",
        "def tokenize_couple(ex):\n",
        "  dct=slow_tokenizer(ex[0], max_length=max_len_txt, padding='max_length')\n",
        "  dct2=slow_tokenizer(ex[1]+\"[SEP]\",add_special_tokens=False,max_length=query_len, padding='max_length')\n",
        "  dct2['token_type_ids']=[1 for i in range(query_len)]\n",
        "  dct['input_ids'].extend(dct2['input_ids'])\n",
        "  dct['token_type_ids'].extend(dct2['token_type_ids'])\n",
        "  dct['attention_mask'].extend(dct2['attention_mask'])\n",
        "  return dct\n",
        "len(tokenize_couple(ex)['input_ids'])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pXa0NV7SFAmm",
        "outputId": "8184d649-c05b-4e47-99b8-95b67e40b1b7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "266"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "rorRlUer-shD"
      },
      "outputs": [],
      "source": [
        "def create_inputs_targets():\n",
        "  input_ids= []\n",
        "  token_type_ids= []\n",
        "  attention_mask= []\n",
        "  tags = []\n",
        "  for i, row in df.iterrows():\n",
        "    dic = tokenize_couple((row['txt'],row['categ_txt']))\n",
        "    tagging = [0]+list(map(lambda x: 0 if x=='O' else 1 if x=='B' else 2,row['tagging']))+[0]\n",
        "    tagging.extend([3]*(max_len_txt-len(tagging)-1))\n",
        "    input_ids.append(dic['input_ids'])\n",
        "    token_type_ids.append(dic['token_type_ids'])\n",
        "    attention_mask.append(dic['attention_mask'])\n",
        "    tags.append(tagging)          \n",
        "  x = [\n",
        "      tf.convert_to_tensor(np.array(input_ids), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(token_type_ids), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(attention_mask), dtype=tf.dtypes.int32),\n",
        "  ]\n",
        "  return x, tf.convert_to_tensor(np.array(tags,dtype=np.dtype('i')),dtype=tf.dtypes.int32)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "7dY5OwkZUXna"
      },
      "outputs": [],
      "source": [
        "x_train, y_train = create_inputs_targets()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "gIUqIaS_32g3"
      },
      "outputs": [],
      "source": [
        "loss_object = tf.keras.losses.SparseCategoricalCrossentropy(\n",
        "    from_logits=False, reduction=tf.keras.losses.Reduction.NONE\n",
        ")\n",
        "\n",
        "\n",
        "def masked_ce_loss(real, pred):\n",
        "    mask = tf.math.logical_not(tf.math.equal(real, 3))\n",
        "    loss_ = loss_object(real, pred)\n",
        "    mask = tf.cast(mask, dtype=loss_.dtype)\n",
        "    loss_ *= mask\n",
        "    return tf.reduce_mean(loss_)\n",
        "\n",
        "def create_model(num_tags):\n",
        "    ## BERT encoder\n",
        "    encoder = TFBertModel.from_pretrained(\"bert-base-uncased\")\n",
        "\n",
        "    ## NER Model\n",
        "    input_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    token_type_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    attention_mask = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    embedding = encoder(\n",
        "        input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask\n",
        "    )['last_hidden_state'][:,:max_len_txt-1,:]\n",
        "    embedding = layers.Dropout(0.2)(embedding)\n",
        "    l = layers.Dense(32,activation=\"tanh\")(embedding)\n",
        "    tag_logits = layers.Dense(num_tags, activation='softmax')(l)\n",
        "    \n",
        "    model = keras.Model(\n",
        "        inputs=[input_ids, token_type_ids, attention_mask],\n",
        "        outputs=[tag_logits],\n",
        "    )\n",
        "    optimizer = keras.optimizers.Adam(learning_rate=3e-5)\n",
        "    model.compile(optimizer=optimizer, loss=masked_ce_loss, metrics=['accuracy'])\n",
        "    return model"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def our_accuracy(y_true, y_pred):\n",
        "  y_true = tf.cast(y_true, tf.int32, name=None)\n",
        "  y_pred = tf.cast(y_pred, tf.int32, name=None)\n",
        "  mask = tf.math.logical_not(tf.math.equal(y_true, 3))\n",
        "  total_no_padding = tf.reduce_sum(tf.cast(mask, tf.int32))\n",
        "  mask = tf.cast(mask, dtype=tf.int32)\n",
        "  # argmax_out = np.argmax(y_pred,-1)\n",
        "  argmax_out = tf.math.argmax(\n",
        "    y_pred,\n",
        "    axis=-1,\n",
        "    output_type=tf.dtypes.int32,\n",
        "    name=None\n",
        "  )\n",
        "  print(y_true)\n",
        "  print(mask)\n",
        "  y_true *= mask\n",
        "  argmax_out *= mask\n",
        "  elements_equal_to_value = tf.equal(y_true, argmax_out)\n",
        "  as_ints = tf.cast(elements_equal_to_value, tf.int32)\n",
        "  count = tf.reduce_sum(as_ints)\n",
        "  res = (count-(tf.size(y_true)-total_no_padding))/total_no_padding\n",
        "  return res"
      ],
      "metadata": {
        "id": "ktGowl4E4L4t"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Q2jAoOvg4_GO"
      },
      "outputs": [],
      "source": [
        "num_tags = 4 # BIOP\n",
        "use_tpu = None\n",
        "try:\n",
        "    tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n",
        "    use_tpu = True\n",
        "except:\n",
        "    use_tpu = False\n",
        "\n",
        "if use_tpu:\n",
        "    # Create distribution strategy\n",
        "    tf.config.experimental_connect_to_cluster(tpu)\n",
        "    tf.tpu.experimental.initialize_tpu_system(tpu)\n",
        "    strategy = tf.distribute.experimental.TPUStrategy(tpu)\n",
        "\n",
        "    # Create model\n",
        "    with strategy.scope():\n",
        "        model = create_model(num_tags)\n",
        "else:\n",
        "    model = create_model(num_tags)\n",
        "    \n",
        "model.summary()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "jXYDVfmR-Mw3",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9a18c89e-b72c-4a6c-aed4-70fa3ea3ba77"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/2\n",
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_3/bert/pooler/dense/kernel:0', 'tf_bert_model_3/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_3/bert/pooler/dense/kernel:0', 'tf_bert_model_3/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_3/bert/pooler/dense/kernel:0', 'tf_bert_model_3/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:tensorflow:Gradients do not exist for variables ['tf_bert_model_3/bert/pooler/dense/kernel:0', 'tf_bert_model_3/bert/pooler/dense/bias:0'] when minimizing the loss. If you're using `model.compile()`, did you forget to provide a `loss`argument?\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "170/170 [==============================] - 126s 167ms/step - loss: nan - accuracy: 0.0678 - val_loss: nan - val_accuracy: 0.0716\n",
            "Epoch 2/2\n",
            "170/170 [==============================] - 16s 93ms/step - loss: nan - accuracy: 0.0689 - val_loss: nan - val_accuracy: 0.0722\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f0c4429cd90>"
            ]
          },
          "metadata": {},
          "execution_count": 93
        }
      ],
      "source": [
        "bs = 16\n",
        "# bs=8\n",
        "\n",
        "model.fit(\n",
        "    x_train,\n",
        "    y_train,\n",
        "    epochs=2,\n",
        "    verbose=1,\n",
        "    batch_size=bs,\n",
        "    validation_split=0.1\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Inference**"
      ],
      "metadata": {
        "id": "c4fvLiKBOtSc"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "3C8NgjHyDWnu",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cd9f6460-ee5a-45e3-c069-8b1ead64454c"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[CLS]-O \n",
            "the-O \n",
            "pizza-O \n",
            "was-O \n",
            "very-O \n",
            "good-O \n",
            "but-O \n",
            "the-O \n",
            "service-B \n",
            "wasn-O \n",
            "'-O \n",
            "t-O \n",
            ".-O \n",
            "[SEP]-O \n"
          ]
        }
      ],
      "source": [
        "c=(\"The pizza was very good but the service wasn't.\",\"food quality negative\")\n",
        "# c=(\"The spaghetti were very tasty.\",\"food general positive\")\n",
        "tokens=tokenize_couple(c)\n",
        "x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "out = model.predict(x)\n",
        "out = np.array(out)\n",
        "argmax_out = np.argmax(out,-1)\n",
        "tag_list = [\"O\",\"B\",\"I\",\"P\"]\n",
        "for i,t in enumerate(slow_tokenizer.tokenize(\"[CLS]\"+c[0]+\"[SEP]\")):\n",
        "  print(\"{}-{} \".format(t,tag_list[argmax_out[0][i]]))"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Save the model**"
      ],
      "metadata": {
        "id": "VdYLSVtF8r_2"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# model.save(\"/content/drive/MyDrive/HLT/ner_model_balanced.h5\")"
      ],
      "metadata": {
        "id": "MOF5Lz0n8rUu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Final model"
      ],
      "metadata": {
        "id": "RXpxaJE7Vq77"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "import re\n",
        "import json\n",
        "import string\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "from sklearn import preprocessing\n",
        "\n",
        "max_len=266\n",
        "query_len=10\n",
        "max_len_txt=max_len-query_len\n",
        "\n",
        "import pandas as pd\n",
        "df = pd.read_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")"
      ],
      "metadata": {
        "id": "UGjwKA9MVuDF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def tokenize_couple(ex):\n",
        "  dct=slow_tokenizer(ex[0], max_length=max_len_txt, padding='max_length')\n",
        "  dct2=slow_tokenizer(ex[1]+\"[SEP]\",add_special_tokens=False,max_length=query_len, padding='max_length')\n",
        "  dct2['token_type_ids']=[1 for i in range(query_len)]\n",
        "  dct['input_ids'].extend(dct2['input_ids'])\n",
        "  dct['token_type_ids'].extend(dct2['token_type_ids'])\n",
        "  dct['attention_mask'].extend(dct2['attention_mask'])\n",
        "  return dct"
      ],
      "metadata": {
        "id": "qFQgGXtZWgmT"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def create_inputs_targets():\n",
        "  input_ids= []\n",
        "  token_type_ids= []\n",
        "  attention_mask= []\n",
        "  tags = []\n",
        "  f = []\n",
        "  for i, row in df.iterrows():\n",
        "    dic = tokenize_couple((row['txt'],row['categ_txt']))\n",
        "    tagging = [0]+list(map(lambda x: 0 if x=='O' else 1 if x=='B' else 2,row['tagging']))+[0]\n",
        "    tagging.extend([3]*(max_len_txt-len(tagging)-1))\n",
        "    input_ids.append(dic['input_ids'])\n",
        "    token_type_ids.append(dic['token_type_ids'])\n",
        "    attention_mask.append(dic['attention_mask'])\n",
        "    tags.append(tagging)\n",
        "    f.append(1 if row['f']=='yes' else 0)          \n",
        "  x = [\n",
        "      tf.convert_to_tensor(np.array(input_ids[100:]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(token_type_ids[100:]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(attention_mask[100:]), dtype=tf.dtypes.int32),\n",
        "  ]\n",
        "  x_val = [      \n",
        "      tf.convert_to_tensor(np.array(input_ids[:100]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(token_type_ids[:100]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(attention_mask[:100]), dtype=tf.dtypes.int32),\n",
        "  ]\n",
        "  return x, tf.convert_to_tensor(np.array(tags[100:],dtype=np.dtype('i')),dtype=tf.dtypes.int32), tf.convert_to_tensor(np.array(f[100:],dtype=np.dtype('i')),dtype=tf.dtypes.int32), x_val, tf.convert_to_tensor(np.array(tags[:100],dtype=np.dtype('i')),dtype=tf.dtypes.int32), tf.convert_to_tensor(np.array(f[:100],dtype=np.dtype('i')),dtype=tf.dtypes.int32)"
      ],
      "metadata": {
        "id": "f6iauKfDWkxv"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_train, y_train_t, y_train_f, x_valid, y_valid_t, y_valid_f = create_inputs_targets()"
      ],
      "metadata": {
        "id": "-bqOUzg6Woox"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "loss_object = tf.keras.losses.SparseCategoricalCrossentropy(\n",
        "    from_logits=False, reduction=tf.keras.losses.Reduction.NONE\n",
        ")\n",
        "\n",
        "def masked_ce_loss(real, pred):\n",
        "    mask = tf.math.logical_not(tf.math.equal(real, 3))\n",
        "    loss_ = loss_object(real, pred)\n",
        "    mask = tf.cast(mask, dtype=loss_.dtype)\n",
        "    loss_ *= mask\n",
        "    return tf.reduce_mean(loss_)\n",
        "\n",
        "def create_model(num_tags):\n",
        "    ## BERT encoder\n",
        "    #encoder\n",
        "    encoder = TFBertModel.from_pretrained(\"bert-base-uncased\")\n",
        "    ## NER Model\n",
        "    input_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    token_type_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    attention_mask = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    out = encoder(\n",
        "        input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask\n",
        "    )\n",
        "    embedding=out['pooler_output']    \n",
        "    embedding = layers.Dropout(0.1)(embedding)\n",
        "    l = layers.Dense(16, activation=\"tanh\")(embedding)\n",
        "    output = layers.Dense(2, activation='softmax',name=\"category_output\")(l)\n",
        "\n",
        "    ner_embedding=out['last_hidden_state'][:,:max_len_txt-1,:]    \n",
        "    ner_embedding = layers.Dropout(0.1)(ner_embedding)\n",
        "    ner_l = layers.Dense(16, activation=\"tanh\")(ner_embedding)\n",
        "    ner_output = layers.Dense(num_tags, activation='softmax',name=\"ner_output\")(ner_l)\n",
        "    \n",
        "    model = keras.Model(\n",
        "        inputs=[input_ids, token_type_ids, attention_mask],\n",
        "        outputs=[output,ner_output],\n",
        "    )\n",
        "\n",
        "    losses = {\n",
        "\t    \"category_output\": tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),\n",
        "\t    \"ner_output\": masked_ce_loss,\n",
        "    }\n",
        "    lossWeights = {\"category_output\": 1.0, \"ner_output\": 1.0}\n",
        "\n",
        "    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0), \n",
        "              loss=losses, loss_weights=lossWeights,\n",
        "              metrics=[tf.keras.metrics.SparseCategoricalAccuracy('accuracy')])\n",
        "\n",
        "    return model"
      ],
      "metadata": {
        "id": "b0BkEX_WWzuX"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "num_tags = 4 # BIOP\n",
        "use_tpu = None\n",
        "try:\n",
        "    tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n",
        "    use_tpu = True\n",
        "except:\n",
        "    use_tpu = False\n",
        "\n",
        "if use_tpu:\n",
        "    # Create distribution strategy\n",
        "    tf.config.experimental_connect_to_cluster(tpu)\n",
        "    tf.tpu.experimental.initialize_tpu_system(tpu)\n",
        "    strategy = tf.distribute.TPUStrategy(tpu)\n",
        "\n",
        "    # Create model\n",
        "    with strategy.scope():\n",
        "        final_model = create_model(num_tags)\n",
        "else:\n",
        "    final_model = create_model(num_tags)\n",
        "    \n",
        "tf.keras.utils.plot_model(final_model)"
      ],
      "metadata": {
        "id": "0cAufvJPbGD4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 356
        },
        "outputId": "f400fdd2-21da-4257-a61d-256df4073b18"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "NameError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-11-dffe1a75fa4b>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     17\u001b[0m         \u001b[0mfinal_model\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcreate_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnum_tags\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     18\u001b[0m \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 19\u001b[0;31m     \u001b[0mfinal_model\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcreate_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnum_tags\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     20\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     21\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mkeras\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mutils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mplot_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfinal_model\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m<ipython-input-10-a163dcef3f54>\u001b[0m in \u001b[0;36mcreate_model\u001b[0;34m(num_tags)\u001b[0m\n\u001b[1;32m     13\u001b[0m     \u001b[0;31m## BERT encoder\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m     \u001b[0;31m#encoder\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 15\u001b[0;31m     \u001b[0mencoder\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mTFBertModel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfrom_pretrained\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"bert-base-uncased\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     16\u001b[0m     \u001b[0;31m## NER Model\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     17\u001b[0m     \u001b[0minput_ids\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mlayers\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mInput\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmax_len\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdtype\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mint32\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mNameError\u001b[0m: name 'TFBertModel' is not defined"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "bs = 16\n",
        "# bs=8\n",
        "\n",
        "final_model.fit(\n",
        "    x=x_train,\n",
        "    y={\"category_output\":y_train_f, \"ner_output\":y_train_t},\n",
        "    # y=y_train_f,\n",
        "    validation_data=(x_valid,\n",
        "\t\t{\"category_output\": y_valid_f, \"ner_output\": y_valid_t}),\n",
        "    epochs=3,\n",
        "    verbose=1,\n",
        "    batch_size=bs\n",
        ")"
      ],
      "metadata": {
        "id": "TU084seXcayf",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "67108920-5b1f-4405-bf91-66c5631ec4c3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/3\n",
            "182/182 [==============================] - 134s 167ms/step - loss: nan - category_output_loss: 0.5525 - ner_output_loss: nan - category_output_accuracy: 0.7160 - ner_output_accuracy: 0.0717 - val_loss: nan - val_category_output_loss: 0.5499 - val_ner_output_loss: nan - val_category_output_accuracy: 0.7500 - val_ner_output_accuracy: 0.0630\n",
            "Epoch 2/3\n",
            "182/182 [==============================] - 18s 99ms/step - loss: nan - category_output_loss: 0.3467 - ner_output_loss: nan - category_output_accuracy: 0.8566 - ner_output_accuracy: 0.0686 - val_loss: nan - val_category_output_loss: 0.4290 - val_ner_output_loss: nan - val_category_output_accuracy: 0.8200 - val_ner_output_accuracy: 0.0631\n",
            "Epoch 3/3\n",
            "182/182 [==============================] - 18s 99ms/step - loss: nan - category_output_loss: 0.2000 - ner_output_loss: nan - category_output_accuracy: 0.9298 - ner_output_accuracy: 0.0687 - val_loss: nan - val_category_output_loss: 0.4548 - val_ner_output_loss: nan - val_category_output_accuracy: 0.8100 - val_ner_output_accuracy: 0.0630\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f1793cc8810>"
            ]
          },
          "metadata": {},
          "execution_count": 37
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#Inference\n",
        "c=(\"The waiter was attentive, the food was delicious and the views of the city were great.\",\"food general positive\")\n",
        "# c=(\"The pasta were very tasty.\",\"food quality positive\")\n",
        "tokens=tokenize_couple(c)\n",
        "x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "out = final_model.predict(x)\n",
        "print(out[0])\n",
        "argmax_out = np.argmax(out[1],-1)\n",
        "tag_list = [\"O\",\"B\",\"I\",\"P\"]\n",
        "for i,t in enumerate(slow_tokenizer.tokenize(\"[CLS]\"+c[0]+\"[SEP]\")):\n",
        "  print(\"{}-{} \".format(t,tag_list[argmax_out[0][i]]))"
      ],
      "metadata": {
        "id": "QzSEMf3fcCEZ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c1e28a49-29ba-4e56-c7a0-a06804bf01b3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[0.02532563 0.9746743 ]]\n",
            "[CLS]-O \n",
            "the-O \n",
            "waiter-O \n",
            "was-O \n",
            "at-O \n",
            "##ten-O \n",
            "##tive-O \n",
            ",-O \n",
            "the-O \n",
            "food-B \n",
            "was-O \n",
            "delicious-O \n",
            "and-O \n",
            "the-O \n",
            "views-O \n",
            "of-O \n",
            "the-O \n",
            "city-O \n",
            "were-O \n",
            "great-O \n",
            ".-O \n",
            "[SEP]-O \n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# ROBERTA"
      ],
      "metadata": {
        "id": "KwACs4DM6fWC"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "import re\n",
        "import json\n",
        "import string\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "from sklearn import preprocessing\n",
        "\n",
        "max_len=200\n",
        "query_len=10\n",
        "max_len_txt=max_len-query_len\n",
        "\n",
        "import pandas as pd\n",
        "df = pd.read_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")"
      ],
      "metadata": {
        "id": "hnD3VsyU6oKH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install transformers\n",
        "from transformers import RobertaTokenizer, TFRobertaModel\n",
        "import tensorflow as tf\n",
        "\n",
        "tokenizer = RobertaTokenizer.from_pretrained(\"roberta-base\")\n",
        "# model = TFRobertaModel.from_pretrained(\"roberta-base\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "66j5_wv-zRM9",
        "outputId": "b05c8902-5435-452f-b6ce-304759fd36e8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/\n",
            "Requirement already satisfied: transformers in /usr/local/lib/python3.7/dist-packages (4.19.2)\n",
            "Requirement already satisfied: numpy>=1.17 in /usr/local/lib/python3.7/dist-packages (from transformers) (1.21.6)\n",
            "Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.7/dist-packages (from transformers) (21.3)\n",
            "Requirement already satisfied: filelock in /usr/local/lib/python3.7/dist-packages (from transformers) (3.7.0)\n",
            "Requirement already satisfied: importlib-metadata in /usr/local/lib/python3.7/dist-packages (from transformers) (4.11.3)\n",
            "Requirement already satisfied: pyyaml>=5.1 in /usr/local/lib/python3.7/dist-packages (from transformers) (6.0)\n",
            "Requirement already satisfied: requests in /usr/local/lib/python3.7/dist-packages (from transformers) (2.23.0)\n",
            "Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.7/dist-packages (from transformers) (2019.12.20)\n",
            "Requirement already satisfied: huggingface-hub<1.0,>=0.1.0 in /usr/local/lib/python3.7/dist-packages (from transformers) (0.7.0)\n",
            "Requirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.7/dist-packages (from transformers) (4.64.0)\n",
            "Requirement already satisfied: tokenizers!=0.11.3,<0.13,>=0.11.1 in /usr/local/lib/python3.7/dist-packages (from transformers) (0.12.1)\n",
            "Requirement already satisfied: typing-extensions>=3.7.4.3 in /usr/local/lib/python3.7/dist-packages (from huggingface-hub<1.0,>=0.1.0->transformers) (4.2.0)\n",
            "Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /usr/local/lib/python3.7/dist-packages (from packaging>=20.0->transformers) (3.0.9)\n",
            "Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.7/dist-packages (from importlib-metadata->transformers) (3.8.0)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (1.24.3)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (3.0.4)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (2.10)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (2022.5.18.1)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def tokenize_couple(ex):\n",
        "  dct=tokenizer(ex[0], max_length=max_len_txt, padding='max_length',add_special_tokens=True)\n",
        "  dct2=tokenizer(\"</s>\"+ex[1]+\"</s>\",add_special_tokens=False,max_length=query_len, padding='max_length')\n",
        "  dct['token_type_ids'] = [0]* max_len_txt\n",
        "  dct2['token_type_ids']=[1 for i in range(query_len)]\n",
        "  dct['input_ids'].extend(dct2['input_ids'])\n",
        "  dct['token_type_ids'].extend(dct2['token_type_ids'])\n",
        "  dct['attention_mask'].extend(dct2['attention_mask'])\n",
        "  return dct"
      ],
      "metadata": {
        "id": "pEkivzdv6tP1"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def create_inputs_targets():\n",
        "  input_ids= []\n",
        "  token_type_ids= []\n",
        "  attention_mask= []\n",
        "  tags = []\n",
        "  f = []\n",
        "  for i, row in df.iterrows():\n",
        "    dic = tokenize_couple((row['txt'],row['categ_txt']))\n",
        "    tagging = [0]+list(map(lambda x: 0 if x=='O' else 1 if x=='B' else 2,row['tagging']))+[0]\n",
        "    tagging.extend([3]*(max_len_txt-len(tagging)-1))\n",
        "    input_ids.append(dic['input_ids'])\n",
        "    token_type_ids.append(dic['token_type_ids'])\n",
        "    attention_mask.append(dic['attention_mask'])\n",
        "    tags.append(tagging)\n",
        "    f.append(1 if row['f']=='yes' else 0)          \n",
        "  x = [\n",
        "      tf.convert_to_tensor(np.array(input_ids[100:]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(token_type_ids[100:]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(attention_mask[100:]), dtype=tf.dtypes.int32),\n",
        "  ]\n",
        "  x_val = [      \n",
        "      tf.convert_to_tensor(np.array(input_ids[:100]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(token_type_ids[:100]), dtype=tf.dtypes.int32),\n",
        "      tf.convert_to_tensor(np.array(attention_mask[:100]), dtype=tf.dtypes.int32),\n",
        "  ]\n",
        "  return x, tf.convert_to_tensor(np.array(tags[100:],dtype=np.dtype('i')),dtype=tf.dtypes.int32), tf.convert_to_tensor(np.array(f[100:],dtype=np.dtype('i')),dtype=tf.dtypes.int32), x_val, tf.convert_to_tensor(np.array(tags[:100],dtype=np.dtype('i')),dtype=tf.dtypes.int32), tf.convert_to_tensor(np.array(f[:100],dtype=np.dtype('i')),dtype=tf.dtypes.int32)"
      ],
      "metadata": {
        "id": "2dqIiFB3DoIJ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_train, y_train_t, y_train_f, x_valid, y_valid_t, y_valid_f = create_inputs_targets()"
      ],
      "metadata": {
        "id": "Wk2S6NVtE7WG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "loss_object = tf.keras.losses.SparseCategoricalCrossentropy(\n",
        "    from_logits=False, reduction=tf.keras.losses.Reduction.NONE\n",
        ")\n",
        "\n",
        "def masked_ce_loss(real, pred):\n",
        "    mask = tf.math.logical_not(tf.math.equal(real, 3))\n",
        "    loss_ = loss_object(real, pred)\n",
        "    mask = tf.cast(mask, dtype=loss_.dtype)\n",
        "    loss_ *= mask\n",
        "    return tf.reduce_mean(loss_)\n",
        "\n",
        "def create_model(num_tags):\n",
        "    ## BERT encoder\n",
        "    #encoder\n",
        "    encoder = TFRobertaModel.from_pretrained(\"roberta-base\")\n",
        "    ## NER Model\n",
        "    input_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    token_type_ids = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    attention_mask = layers.Input(shape=(max_len,), dtype=tf.int32)\n",
        "    out = encoder(\n",
        "        input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask\n",
        "    )\n",
        "    embedding=out['pooler_output']    \n",
        "    embedding = layers.Dropout(0.1)(embedding)\n",
        "    l = layers.Dense(32, activation=\"tanh\")(embedding)\n",
        "    output = layers.Dense(2, activation='softmax',name=\"category_output\")(l)\n",
        "\n",
        "    ner_embedding=out['last_hidden_state'][:,:max_len_txt-1,:]    \n",
        "    ner_embedding = layers.Dropout(0.1)(ner_embedding)\n",
        "    # ner_l = layers.Dense(32, activation=\"tanh\")(ner_embedding)\n",
        "    ner_output = layers.Dense(num_tags, activation='softmax',name=\"ner_output\")(ner_embedding)\n",
        "    \n",
        "    model = keras.Model(\n",
        "        inputs=[input_ids, token_type_ids, attention_mask],\n",
        "        outputs=[output,ner_output],\n",
        "    )\n",
        "\n",
        "    losses = {\n",
        "\t    \"category_output\": tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),\n",
        "\t    \"ner_output\": masked_ce_loss,\n",
        "    }\n",
        "    lossWeights = {\"category_output\": 1.0, \"ner_output\": 1.0}\n",
        "\n",
        "    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0), \n",
        "              loss=losses, loss_weights=lossWeights,\n",
        "              metrics=[tf.keras.metrics.SparseCategoricalAccuracy('accuracy')])\n",
        "\n",
        "    return model"
      ],
      "metadata": {
        "id": "gOtyNc3IFLC8"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "num_tags = 4 # BIOP\n",
        "use_tpu = None\n",
        "try:\n",
        "    tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n",
        "    use_tpu = True\n",
        "except:\n",
        "    use_tpu = False\n",
        "\n",
        "if use_tpu:\n",
        "    # Create distribution strategy\n",
        "    tf.config.experimental_connect_to_cluster(tpu)\n",
        "    tf.tpu.experimental.initialize_tpu_system(tpu)\n",
        "    strategy = tf.distribute.TPUStrategy(tpu)\n",
        "\n",
        "    # Create model\n",
        "    with strategy.scope():\n",
        "        final_model = create_model(num_tags)\n",
        "else:\n",
        "    final_model = create_model(num_tags)\n",
        "    \n",
        "tf.keras.utils.plot_model(final_model)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "fk6OqQGUFjPj",
        "outputId": "def32418-185b-4548-c9bc-ea5d24539677"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Deallocate tpu buffers before initializing tpu system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Deallocate tpu buffers before initializing tpu system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:TPU system grpc://10.123.7.210:8470 has already been initialized. Reinitializing the TPU can cause previously created variables on TPU to be lost.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "WARNING:tensorflow:TPU system grpc://10.123.7.210:8470 has already been initialized. Reinitializing the TPU can cause previously created variables on TPU to be lost.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Initializing the TPU system: grpc://10.123.7.210:8470\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Initializing the TPU system: grpc://10.123.7.210:8470\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Finished initializing TPU system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Finished initializing TPU system.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Found TPU system:\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:Found TPU system:\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Num TPU Workers: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Num TPU Workers: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores Per Worker: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Num TPU Cores Per Worker: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:localhost/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:localhost/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:CPU:0, CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:0, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:0, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:1, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:1, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:2, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:2, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:3, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:3, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:4, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:4, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:5, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:5, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:6, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:6, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:7, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU:7, TPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU_SYSTEM:0, TPU_SYSTEM, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:TPU_SYSTEM:0, TPU_SYSTEM, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:XLA_CPU:0, XLA_CPU, 0, 0)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "INFO:tensorflow:*** Available Device: _DeviceAttributes(/job:worker/replica:0/task:0/device:XLA_CPU:0, XLA_CPU, 0, 0)\n",
            "Some layers from the model checkpoint at roberta-base were not used when initializing TFRobertaModel: ['lm_head']\n",
            "- This IS expected if you are initializing TFRobertaModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing TFRobertaModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "All the layers of TFRobertaModel were initialized from the model checkpoint at roberta-base.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFRobertaModel for predictions without further training.\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<IPython.core.display.Image object>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 33
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "bs = 8\n",
        "# bs=8\n",
        "\n",
        "final_model.fit(\n",
        "    x=x_train,\n",
        "    y={\"category_output\":y_train_f, \"ner_output\":y_train_t},\n",
        "    # y=y_train_f,\n",
        "    validation_data=(x_valid,\n",
        "\t\t{\"category_output\": y_valid_f, \"ner_output\": y_valid_t}),\n",
        "    epochs=3,\n",
        "    verbose=1,\n",
        "    batch_size=bs\n",
        ")"
      ],
      "metadata": {
        "id": "aGc67kAaFqR9",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7f099c28-ac84-43cb-9c63-523cd228f0b8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/3\n",
            "364/364 [==============================] - 33s 92ms/step - loss: nan - category_output_loss: 0.6943 - ner_output_loss: nan - category_output_accuracy: 0.5062 - ner_output_accuracy: 0.0925 - val_loss: nan - val_category_output_loss: 0.6936 - val_ner_output_loss: nan - val_category_output_accuracy: 0.4500 - val_ner_output_accuracy: 0.0850\n",
            "Epoch 2/3\n",
            "364/364 [==============================] - 33s 92ms/step - loss: nan - category_output_loss: 0.6936 - ner_output_loss: nan - category_output_accuracy: 0.5072 - ner_output_accuracy: 0.0925 - val_loss: nan - val_category_output_loss: 0.6940 - val_ner_output_loss: nan - val_category_output_accuracy: 0.4500 - val_ner_output_accuracy: 0.0850\n",
            "Epoch 3/3\n",
            "364/364 [==============================] - 34s 94ms/step - loss: nan - category_output_loss: 0.6938 - ner_output_loss: nan - category_output_accuracy: 0.5010 - ner_output_accuracy: 0.0925 - val_loss: nan - val_category_output_loss: 0.6916 - val_ner_output_loss: nan - val_category_output_accuracy: 0.5500 - val_ner_output_accuracy: 0.0850\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f2356543810>"
            ]
          },
          "metadata": {},
          "execution_count": 36
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# final_model.save(\"/content/drive/MyDrive/HLT/jointed_roberta.h5\")\n",
        "# final_model.load_weights(\"/content/drive/MyDrive/HLT/jointed_roberta.h5\")"
      ],
      "metadata": {
        "id": "MT_Q_yoaPGlo"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#Inference\n",
        "c=(\"The waiter was attentive, the food was delicious and the views of the city were great.\",\"food quality positive\")\n",
        "# c=(\"The pasta were very tasty.\",\"food quality positive\")\n",
        "#c = (\"There is something about their atmosphere that makes me come back nearly every week.\",\"ambience general positive\")\n",
        "tokens=tokenize_couple(c)\n",
        "x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "out = final_model.predict(x)\n",
        "print(out[0])\n",
        "argmax_out = np.argmax(out[1],-1)\n",
        "tag_list = [\"O\",\"B\",\"I\",\"P\"]\n",
        "for i,t in enumerate(tokenizer.tokenize(\"<s>\"+c[0])):\n",
        "  print(\"{}-{} \".format(t.replace(\"Ġ\",\"\"),tag_list[argmax_out[0][i]]))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "0kzHCUJEFufy",
        "outputId": "133bafab-7922-49de-eb64-2b796e896f26"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[0.4851598  0.51484025]]\n",
            "<s>-O \n",
            "The-O \n",
            "waiter-O \n",
            "was-O \n",
            "attentive-O \n",
            ",-O \n",
            "the-O \n",
            "food-O \n",
            "was-O \n",
            "delicious-O \n",
            "and-O \n",
            "the-O \n",
            "views-O \n",
            "of-O \n",
            "the-O \n",
            "city-O \n",
            "were-O \n",
            "great-O \n",
            ".-O \n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import pickle\n",
        "pkl_file = open('/content/drive/MyDrive/HLT/cat_pol_couples.pkl', 'rb')\n",
        "cat_pol_couples = pickle.load(pkl_file)\n",
        "pkl_file.close()"
      ],
      "metadata": {
        "id": "G6VFINUQNbkH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from functools import reduce\n",
        "\n",
        "def inference_function(c,th):\n",
        "  res=[]\n",
        "  tokens=tokenize_couple(c)\n",
        "  x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "  out = final_model.predict(x,verbose=0,)\n",
        "  out1 = out[0]\n",
        "  out2 = out[1]\n",
        "  out1[out1 < th] = 0 #thresholding.\n",
        "  indice = np.argmax(out1[0])\n",
        "  classi=[\"no\",\"yes\"]\n",
        "  # print(classi[indice])\n",
        "  tag_couple = []\n",
        "  category = \"#\".join(c[1].split(\" \")[:-1])\n",
        "  category = category.upper()\n",
        "  if indice==1:\n",
        "    argmax_out = np.argmax(out2,-1)\n",
        "    if 1 in argmax_out:\n",
        "      tag_list = [\"O\",\"B\",\"I\",\"P\"]\n",
        "      tokens = tokenizer.tokenize(\"<s>\"+c[0])\n",
        "      tokens=list(map(lambda x:x.replace('Ġ',''),tokens))\n",
        "      for i,t in enumerate(tokens):\n",
        "        tag_couple.append((t,tag_list[argmax_out[0][i]]))\n",
        "        # print(\"{}-{}\".format(t,tag_list[argmax_out[0][i]]))\n",
        "      \n",
        "      tokens = tokens[1:] #remove [CLS] and [SEP] tokens\n",
        "      list_indices = []\n",
        "      start=0\n",
        "      end=0\n",
        "      previous = 'O'\n",
        "      trovato = False\n",
        "      for i,(t,tag) in enumerate(tag_couple[1:]):\n",
        "        if tag!='O' and not trovato:\n",
        "          start=i\n",
        "          trovato=True\n",
        "        if trovato and tag=='O':\n",
        "          end=i\n",
        "          trovato=False\n",
        "          list_indices.append((start,end))\n",
        "      special_caracters=[':','-','$','[',']',')','(','.',',','?','!']\n",
        "      for z in list_indices:\n",
        "        parte_asx = tokens[:z[0]]\n",
        "        if len(parte_asx)==0:\n",
        "          from_= 0\n",
        "          fake_space_sx = 0\n",
        "        else:\n",
        "          from_ = reduce(lambda x,y:x+y,(map(lambda x:len(x.replace(\"#\",\"\")),parte_asx)))\n",
        "          fake_space_sx = reduce(lambda x,y:x+y,(map(lambda x: 1 if (\"#\" in x) or (x in special_caracters) else 0,parte_asx)))\n",
        "        space_sx = z[0]\n",
        "        from_+=space_sx-fake_space_sx\n",
        "        parte_adx = tokens[:z[1]]\n",
        "        to = reduce(lambda x,y:x+y,(map(lambda x:len(x.replace(\"#\",\"\")),parte_adx)))\n",
        "        space_dx = z[1]\n",
        "        fake_space_dx = reduce(lambda x,y:x+y,(map(lambda x: 1 if (\"#\" in x) or (x in special_caracters) else 0,parte_adx)))\n",
        "        to+=space_dx-fake_space_dx-1\n",
        "        # print(\"target={} category={} polarity={} from:{} to:{}\".format(c[0][from_:to],\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],from_,to))\n",
        "        \n",
        "        res.append({'target':c[0][from_:to], 'category':category, 'polarity':c[1].split(\" \")[-1], 'from':str(from_), 'to':str(to)})\n",
        "        # res+=\"<a style=\\\"font-size:13px\\\">target={} category={} polarity={} from:{} to:{}</a><br>\".format(c[0][from_:to],\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],from_,to)\n",
        "    else:\n",
        "      # print(\"target={} category={} polarity={} from:{} to:{}\".format(\"NULL\",\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],0,0))\n",
        "      # res += \"target={} category={} polarity={} from:{} to:{}\".format(\"NULL\",\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],0,0)\n",
        "      res.append({'target':\"NULL\", 'category':category, 'polarity':c[1].split(\" \")[-1], 'from':'0', 'to':'0'})\n",
        "      # res+=\"<a style=\\\"font-size:13px\\\">target={} category={} polarity={} from:{} to:{}</a><br>\".format(\"NULL\",\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],0,0)\n",
        "  return res\n"
      ],
      "metadata": {
        "id": "D_GF3MpYNkm0"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "inference_function((\"The waiter was attentive, the food was delicious and the views of the city were great.\",\"location quality positive\"),0.90)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3RHW17LzOJVC",
        "outputId": "e463b086-a6b5-4d10-9667-f160472d8b4a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[{'category': 'LOCATION#QUALITY',\n",
              "  'from': '4',\n",
              "  'polarity': 'positive',\n",
              "  'target': 'waiter',\n",
              "  'to': '10'},\n",
              " {'category': 'LOCATION#QUALITY',\n",
              "  'from': '57',\n",
              "  'polarity': 'positive',\n",
              "  'target': 'views',\n",
              "  'to': '62'}]"
            ]
          },
          "metadata": {},
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def big_inference(txt,th=0.90):\n",
        "  res=[]\n",
        "  for cat in cat_pol_couples:\n",
        "    c = (txt,\" \".join(cat[0].lower().split(\"#\"))+\" \"+cat[1])\n",
        "    res += inference_function(c,th)\n",
        "  return res"
      ],
      "metadata": {
        "id": "vhKbuusKPzdN"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# big_inference(\"The waiter was attentive, the food was delicious and the views of the city were great.\")"
      ],
      "metadata": {
        "id": "vI612l6LP0gG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def get_opinions(txt, th=0.90):\n",
        "  predict = [];\n",
        "  \n",
        "  for cat in cat_pol_couples:\n",
        "    c = (txt,\" \".join(cat[0].lower().split(\"#\"))+\" \"+cat[1])\n",
        "    inf = inference_function(c,th)\n",
        "    if len(inf)>0:\n",
        "      predict.extend(inf)\n",
        "\n",
        "  return predict"
      ],
      "metadata": {
        "id": "-Ss5O7GsdGay"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "\n",
        "def jaccardIndex(predict, target):\n",
        "  predict = list(map(str,predict))\n",
        "  target = list(map(str,target))\n",
        "  predict = set(predict)\n",
        "  target = set(target)\n",
        "  \n",
        "  inter = len(predict.intersection(target))\n",
        "  union = len(predict.union(target))\n",
        "  if union==0:\n",
        "    return 1\n",
        "  return inter / union\n",
        "\n",
        "\n",
        "def precision(predict, target):\n",
        "  predict = list(map(str,predict))\n",
        "  target = list(map(str,target))\n",
        "  predict = set(predict)\n",
        "  target = set(target)\n",
        "  \n",
        "  inter = len(predict.intersection(target))\n",
        "  num_predict = len(predict)\n",
        "  if num_predict==0:\n",
        "    return 1\n",
        "  return inter / num_predict\n",
        "\n",
        "def recall(predict, target):\n",
        "  predict = list(map(str,predict))\n",
        "  target = list(map(str,target))\n",
        "  predict = set(predict)\n",
        "  target = set(target)\n",
        "  \n",
        "  inter = len(predict.intersection(target))\n",
        "  num_target = len(target)\n",
        "  if num_target==0:\n",
        "    return 1\n",
        "  return inter / num_target\n",
        "\n",
        "def f1score(predict, target):\n",
        "  p=precision(predict, target)\n",
        "  r=recall(predict, target)\n",
        "  return (2 * p * r) / (p + r) if (p+r)>0 else 0\n",
        "\n",
        "import zipfile\n",
        "import pandas as pd\n",
        "import xml.etree.ElementTree as ET\n",
        "\n",
        "dataset_path = \"/content/drive/MyDrive/HLT/RestaurantTS.zip\"\n",
        "with zipfile.ZipFile(dataset_path,\"r\") as zip_ref:\n",
        "  zip_ref.extractall(\"./dataset\")"
      ],
      "metadata": {
        "id": "6Nh7MqMndKIx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "\n",
        "def evaluate_restaurants():\n",
        "  \"\"\"\n",
        "    Mean of Jaccard index over the sentences\n",
        "  \"\"\"\n",
        "\n",
        "  xml_path = \"/content/dataset/ABSA15_Restaurants_Test.xml\"\n",
        "  mytree = ET.parse(xml_path)\n",
        "  reviews = mytree.getroot()\n",
        "  input = []\n",
        "  all_opinions = []\n",
        "  for rew in reviews:\n",
        "    for sent in rew[0]: #cicle on single sentence in review.\n",
        "      txt = sent[0].text\n",
        "      if len(sent)>1:\n",
        "        input.append(txt)\n",
        "        sent_opinions = []\n",
        "        for opinion in sent[1]:\n",
        "          op_dict = opinion.attrib\n",
        "          sent_opinions.append(op_dict)\n",
        "        all_opinions.append(sent_opinions)\n",
        "\n",
        "  MJI = 0\n",
        "  MPR = 0\n",
        "  MRC = 0\n",
        "  MF1 = 0\n",
        "  tot = len(input)\n",
        "  i = 0\n",
        "  for sent_txt, opinions_target in zip(input, all_opinions):\n",
        "    predict = get_opinions(sent_txt)\n",
        "    MJI += jaccardIndex(predict, opinions_target)\n",
        "    MPR += precision(predict, opinions_target)\n",
        "    MRC += recall(predict, opinions_target)\n",
        "    MF1 += f1score(predict, opinions_target)\n",
        "    i += 1\n",
        "    print(\"Observed={}\".format(i/tot))\n",
        "    print(\"Jaccard={} Precision={} Recall={} F1-score={}\".format(MJI/i,MPR/i,MRC/i,MF1/i))\n",
        "\n",
        "  return MJI/tot, MPR/tot, MRC/tot, MF1/tot"
      ],
      "metadata": {
        "id": "RT7PZ0yMd3go"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "j,p,r,f1 = evaluate_restaurants()\n",
        "print(\"Jaccard:{} Precision:{} Recall:{} F1:{}\".format(j,p,r,f1))"
      ],
      "metadata": {
        "id": "cBjGaMbyeMph"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Inference of models"
      ],
      "metadata": {
        "id": "BtJhqPqWZ5CK"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import pickle\n",
        "\n",
        "df = pd.read_pickle(\"/content/drive/MyDrive/HLT/ds.pkl\")\n",
        "pkl_file = open('/content/drive/MyDrive/HLT/cat_pol_couples.pkl', 'rb')\n",
        "cat_pol_couples = pickle.load(pkl_file)\n",
        "pkl_file.close()"
      ],
      "metadata": {
        "id": "aNoAodDpvY7_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "epochs=1\n",
        "model = create_model_class()\n",
        "model.load_weights(\"/content/drive/MyDrive/HLT/class_model_balanced.h5\")"
      ],
      "metadata": {
        "id": "OazT4j1gZ4iT",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 156,
          "referenced_widgets": [
            "10a4469cbeb540ff8d961d23b1dc58df",
            "95d064bf3eb246d2861fa0a9dfb19ed9",
            "76a8eb371869447488dcc44d8cd315b1",
            "28aa9491c51e4d63994e4c165dff661f",
            "792e7c63bdf248c89ff9ef615752867e",
            "00a41a671d3c4d39927d3c88dd2f4a36",
            "dec923fa743f4f46b4f4644702c43fc2",
            "1cad7d52892e46a6af14a1f6393faafc",
            "c6988e819ce54ad99a847e7e8ada2706",
            "fdfbfd3a5dae45ce89c91eac4870a009",
            "59652c3a3a37432b8c05c52f00af132f"
          ]
        },
        "outputId": "817b7ce4-13ea-4639-9041-6950d07bdc0c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Downloading:   0%|          | 0.00/511M [00:00<?, ?B/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "10a4469cbeb540ff8d961d23b1dc58df"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Some layers from the model checkpoint at bert-base-uncased were not used when initializing TFBertModel: ['nsp___cls', 'mlm___cls']\n",
            "- This IS expected if you are initializing TFBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing TFBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "All the layers of TFBertModel were initialized from the model checkpoint at bert-base-uncased.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFBertModel for predictions without further training.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "ner_model = create_model(4)\n",
        "ner_model.load_weights(\"/content/drive/MyDrive/HLT/ner_model_balanced.h5\")"
      ],
      "metadata": {
        "id": "IsZTyFVebMVi",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a2a246bf-7c35-4c66-83f2-39f4fdff4cb8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Some layers from the model checkpoint at bert-base-uncased were not used when initializing TFBertModel: ['nsp___cls', 'mlm___cls']\n",
            "- This IS expected if you are initializing TFBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing TFBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "All the layers of TFBertModel were initialized from the model checkpoint at bert-base-uncased.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFBertModel for predictions without further training.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from functools import reduce\n",
        "\n",
        "def inference_function(c,th):\n",
        "  res=[]\n",
        "  tokens=tokenize_couple(c)\n",
        "  x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "  out2 = ner_model.predict(x,verbose=0,)\n",
        "  out1 = model.predict(x,verbose=0,)\n",
        "  out1[out1 < th] = 0 #thresholding.\n",
        "  indice = np.argmax(out1[0])\n",
        "  classi=[\"no\",\"yes\"]\n",
        "  # print(classi[indice])\n",
        "  tag_couple = []\n",
        "  category = \"#\".join(c[1].split(\" \")[:-1])\n",
        "  category = category.upper()\n",
        "  if indice==1:\n",
        "    argmax_out = np.argmax(out2,-1)\n",
        "    if 1 in argmax_out:\n",
        "      tag_list = [\"O\",\"B\",\"I\",\"P\"]\n",
        "      tokens = slow_tokenizer.tokenize(\"[CLS]\"+c[0]+\"[SEP]\")\n",
        "      for i,t in enumerate(tokens):\n",
        "        tag_couple.append((t,tag_list[argmax_out[0][i]]))\n",
        "        # print(\"{}-{}\".format(t,tag_list[argmax_out[0][i]]))\n",
        "      \n",
        "      tokens = tokens[1:-1] #remove [CLS] and [SEP] tokens\n",
        "      list_indices = []\n",
        "      start=0\n",
        "      end=0\n",
        "      previous = 'O'\n",
        "      trovato = False\n",
        "\n",
        "      for i,(t,tag) in enumerate(tag_couple[1:-1]):\n",
        "        if tag!='O' and not trovato:\n",
        "          start=i\n",
        "          trovato=True\n",
        "        if trovato and tag=='O':\n",
        "          end=i\n",
        "          trovato=False\n",
        "          list_indices.append((start,end))\n",
        "      special_caracters=[':','-','$','[',']',')','(','.',',','?','!']\n",
        "      for z in list_indices:\n",
        "        parte_asx = tokens[:z[0]]\n",
        "        if len(parte_asx)==0:\n",
        "          from_= 0\n",
        "          fake_space_sx = 0\n",
        "        else:\n",
        "          from_ = reduce(lambda x,y:x+y,(map(lambda x:len(x.replace(\"#\",\"\")),parte_asx)))\n",
        "          fake_space_sx = reduce(lambda x,y:x+y,(map(lambda x: 1 if (\"#\" in x) or (x in special_caracters) else 0,parte_asx)))\n",
        "        space_sx = z[0]\n",
        "        from_+=space_sx-fake_space_sx\n",
        "        parte_adx = tokens[:z[1]]\n",
        "        to = reduce(lambda x,y:x+y,(map(lambda x:len(x.replace(\"#\",\"\")),parte_adx)))\n",
        "        space_dx = z[1]\n",
        "        fake_space_dx = reduce(lambda x,y:x+y,(map(lambda x: 1 if (\"#\" in x) or (x in special_caracters) else 0,parte_adx)))\n",
        "        to+=space_dx-fake_space_dx-1\n",
        "        # print(\"target={} category={} polarity={} from:{} to:{}\".format(c[0][from_:to],\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],from_,to))\n",
        "        \n",
        "        res.append({'target':c[0][from_:to], 'category':category, 'polarity':c[1].split(\" \")[-1], 'from':str(from_), 'to':str(to)})\n",
        "        # res+=\"<a style=\\\"font-size:13px\\\">target={} category={} polarity={} from:{} to:{}</a><br>\".format(c[0][from_:to],\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],from_,to)\n",
        "    else:\n",
        "      # print(\"target={} category={} polarity={} from:{} to:{}\".format(\"NULL\",\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],0,0))\n",
        "      # res += \"target={} category={} polarity={} from:{} to:{}\".format(\"NULL\",\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],0,0)\n",
        "      res.append({'target':\"NULL\", 'category':category, 'polarity':c[1].split(\" \")[-1], 'from':'0', 'to':'0'})\n",
        "      # res+=\"<a style=\\\"font-size:13px\\\">target={} category={} polarity={} from:{} to:{}</a><br>\".format(\"NULL\",\"#\".join(c[1].split(\" \")[:-1]),c[1].split(\" \")[-1],0,0)\n",
        "  return res\n"
      ],
      "metadata": {
        "id": "HPg9dL99dZiZ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "c=(\"I LOOOVE their eggplant pizza, as well as their pastas and their risotti!\",\"food quality positive\")\n",
        "# c = (\"Good food: my favorite is the seafood spaghetti.\",\"food quality positive\")\n",
        "inference_function(c,0.90)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7BUvm9XotCOK",
        "outputId": "d5134c39-381c-4b60-cae9-5fd4f9de337d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[{'category': 'FOOD#QUALITY',\n",
              "  'from': '15',\n",
              "  'polarity': 'positive',\n",
              "  'target': 'eggplant pizza',\n",
              "  'to': '29'},\n",
              " {'category': 'FOOD#QUALITY',\n",
              "  'from': '48',\n",
              "  'polarity': 'positive',\n",
              "  'target': 'pastas',\n",
              "  'to': '54'},\n",
              " {'category': 'FOOD#QUALITY',\n",
              "  'from': '65',\n",
              "  'polarity': 'positive',\n",
              "  'target': 'risotti',\n",
              "  'to': '72'}]"
            ]
          },
          "metadata": {},
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tokenize_couple((\"I love the pizza here!\",\"food quality positive\"))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Ew5EQSH-ReF5",
        "outputId": "bb93e57c-3b6d-489b-a6e9-4ad89ed77a14"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'input_ids': [101, 1045, 2293, 1996, 10733, 2182, 999, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2833, 3737, 3893, 102, 0, 0, 0, 0, 0, 0], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0]}"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def big_inference(txt,th=0.90):\n",
        "  res=\"\"\n",
        "  for cat in cat_pol_couples:\n",
        "    c = (txt,\" \".join(cat[0].lower().split(\"#\"))+\" \"+cat[1])\n",
        "    res += inference_function(c,th)\n",
        "  return res"
      ],
      "metadata": {
        "id": "_6z0S7Qxy5pD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "a = big_inference(\"The waiter was attentive, the food was delicious and the views of the city were great.\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "iLlgFPfKzpEb",
        "outputId": "550299f1-d61a-4ff4-86db-18ff8ed7f907"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "target=food category=food#general polarity=positive from:30 to:34\n",
            "target=waiter category=service#general polarity=positive from:4 to:10\n",
            "target=food category=food#quality polarity=positive from:30 to:34\n",
            "target=views category=ambience#general polarity=positive from:57 to:62\n",
            "target=NULL category=location#general polarity=positive from:0 to:0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "a"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 70
        },
        "id": "toqNP6_Nyw4b",
        "outputId": "99fbddc8-0b86-4a0a-f1bf-0757069de74a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'target=food category=food#general polarity=positive from:30 to:34 \\ntarget=waiter category=service#general polarity=positive from:4 to:10 \\ntarget=food category=food#quality polarity=positive from:30 to:34 \\ntarget=views category=ambience#general polarity=positive from:57 to:62 \\ntarget=NULL category=location#general polarity=positive from:0 to:0 \\n'"
            ],
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            }
          },
          "metadata": {},
          "execution_count": 65
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Gradio"
      ],
      "metadata": {
        "id": "zzzzWSjepudC"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install gradio"
      ],
      "metadata": {
        "id": "hDtx2gSIp7tU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import gradio as gr\n",
        "demo = gr.Interface(\n",
        "    fn=big_inference,\n",
        "    inputs=[\"text\"],\n",
        "    outputs=[\"html\"],\n",
        ")\n",
        "# gr.Slider(0, 1)\n",
        "demo.launch(debug=True)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 747
        },
        "id": "Hs2NTEqzpslp",
        "outputId": "037ef2ed-97d0-4c3b-ff09-997259359956"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Colab notebook detected. This cell will run indefinitely so that you can see errors and logs. To turn off, set debug=False in launch().\n",
            "Running on public URL: https://42923.gradio.app\n",
            "\n",
            "This share link expires in 72 hours. For free permanent hosting, check out Spaces (https://huggingface.co/spaces)\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<IPython.lib.display.IFrame at 0x7fc277df5c10>"
            ],
            "text/html": [
              "\n",
              "        <iframe\n",
              "            width=\"900\"\n",
              "            height=\"500\"\n",
              "            src=\"https://42923.gradio.app\"\n",
              "            frameborder=\"0\"\n",
              "            allowfullscreen\n",
              "        ></iframe>\n",
              "        "
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "target=food category=food#general polarity=positive from:30 to:34\n",
            "target=waiter category=service#general polarity=positive from:4 to:10\n",
            "target=food category=food#quality polarity=positive from:30 to:34\n",
            "target=views category=ambience#general polarity=positive from:57 to:62\n",
            "target=NULL category=location#general polarity=positive from:0 to:0\n",
            "Keyboard interruption in main thread... closing server.\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(<gradio.routes.App at 0x7fc27f33aed0>,\n",
              " 'http://127.0.0.1:7861/',\n",
              " 'https://42923.gradio.app')"
            ]
          },
          "metadata": {},
          "execution_count": 80
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Evaluation"
      ],
      "metadata": {
        "id": "angaujb2S4YW"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from functools import reduce\n",
        "\n",
        "def inference_function(c,th):\n",
        "  res=[]\n",
        "  tokens=tokenize_couple(c)\n",
        "  x = [np.array([tokens['input_ids']]), np.array([tokens['token_type_ids']]), np.array([tokens['attention_mask']])]\n",
        "  out2 = ner_model.predict(x,verbose=0,)\n",
        "  out1 = model.predict(x,verbose=0,)\n",
        "  out1[out1 < th] = 0 #thresholding.\n",
        "  indice = np.argmax(out1[0])\n",
        "  classi=[\"no\",\"yes\"]\n",
        "  tag_couple = []\n",
        "  category = \"#\".join(c[1].split(\" \")[:-1])\n",
        "  category = category.upper()\n",
        "  if indice==1:\n",
        "    argmax_out = np.argmax(out2,-1)\n",
        "    if 1 in argmax_out:\n",
        "      tag_list = [\"O\",\"B\",\"I\",\"P\"]\n",
        "      tokens = slow_tokenizer.tokenize(\"[CLS]\"+c[0]+\"[SEP]\")\n",
        "      for i,t in enumerate(tokens):\n",
        "        tag_couple.append((t,tag_list[argmax_out[0][i]]))\n",
        "      \n",
        "      tokens = tokens[1:-1] #remove [CLS] and [SEP] tokens\n",
        "      list_indices = []\n",
        "      start=0\n",
        "      end=0\n",
        "      previous = 'O'\n",
        "      trovato = False\n",
        "\n",
        "      for i,(t,tag) in enumerate(tag_couple[1:-1]):\n",
        "        if tag!='O' and not trovato:\n",
        "          start=i\n",
        "          trovato=True\n",
        "        if trovato and tag=='O':\n",
        "          end=i\n",
        "          trovato=False\n",
        "          list_indices.append((start,end))\n",
        "      special_caracters=[':','-','$','[',']',')','(','.',',','?','!']\n",
        "      for z in list_indices:\n",
        "        parte_asx = tokens[:z[0]]\n",
        "        if len(parte_asx)==0:\n",
        "          from_= 0\n",
        "          fake_space_sx = 0\n",
        "        else:\n",
        "          from_ = reduce(lambda x,y:x+y,(map(lambda x:len(x.replace(\"#\",\"\")),parte_asx)))\n",
        "          fake_space_sx = reduce(lambda x,y:x+y,(map(lambda x: 1 if (\"#\" in x) or (x in special_caracters) else 0,parte_asx)))\n",
        "        space_sx = z[0]\n",
        "        from_+=space_sx-fake_space_sx\n",
        "        parte_adx = tokens[:z[1]]\n",
        "        to = reduce(lambda x,y:x+y,(map(lambda x:len(x.replace(\"#\",\"\")),parte_adx)))\n",
        "        space_dx = z[1]\n",
        "        fake_space_dx = reduce(lambda x,y:x+y,(map(lambda x: 1 if (\"#\" in x) or (x in special_caracters) else 0,parte_adx)))\n",
        "        to+=space_dx-fake_space_dx-1\n",
        "        \n",
        "        res.append({'target':c[0][from_:to], 'category':category, 'polarity':c[1].split(\" \")[-1], 'from':str(from_), 'to':str(to)})\n",
        "    else:\n",
        "\n",
        "      res.append({'target':\"NULL\", 'category':category, 'polarity':c[1].split(\" \")[-1], 'from':'0', 'to':'0'})\n",
        "  return res"
      ],
      "metadata": {
        "id": "ShL8HDCr2xjW"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def get_opinions(txt, th=0.90):\n",
        "  predict = [];\n",
        "  \n",
        "  for cat in cat_pol_couples:\n",
        "    c = (txt,\" \".join(cat[0].lower().split(\"#\"))+\" \"+cat[1])\n",
        "    inf = inference_function(c,th)\n",
        "    if len(inf)>0:\n",
        "      predict.extend(inf)\n",
        "\n",
        "  return predict"
      ],
      "metadata": {
        "id": "Zgaz_AoomO6B"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import nltk\n",
        "nltk.download('punkt')\n",
        "\n",
        "from nltk.tokenize import sent_tokenize\n",
        "\n",
        "def get_opinions_rawText(txt, th=0.90):\n",
        "  ss = sent_tokenize(txt)\n",
        "\n",
        "  predict = [];\n",
        "  for s in ss:\n",
        "    predict.append(get_opinions(s, th));\n",
        "\n",
        "  return predict"
      ],
      "metadata": {
        "id": "u5z0gjmzS7vl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "065e2d0e-a875-46e9-ffa5-b7a47c044381"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "get_opinions_rawText(\"Hi, i'm chiara pugliese\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 321
        },
        "id": "1hth2OFj3HNZ",
        "outputId": "67262e2e-7e7f-4de4-ece5-fd760215c521"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "NameError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-49-d4c2eaaec45a>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mget_opinions_rawText\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"Hi, i'm chiara pugliese\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m<ipython-input-45-6e3dada5c0a5>\u001b[0m in \u001b[0;36mget_opinions_rawText\u001b[0;34m(txt, th)\u001b[0m\n\u001b[1;32m      9\u001b[0m   \u001b[0mpredict\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m;\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     10\u001b[0m   \u001b[0;32mfor\u001b[0m \u001b[0ms\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mss\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 11\u001b[0;31m     \u001b[0mpredict\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mappend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mget_opinions\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0ms\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mth\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m;\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     12\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     13\u001b[0m   \u001b[0;32mreturn\u001b[0m \u001b[0mpredict\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m<ipython-input-44-f660a7aab58d>\u001b[0m in \u001b[0;36mget_opinions\u001b[0;34m(txt, th)\u001b[0m\n\u001b[1;32m      4\u001b[0m   \u001b[0;32mfor\u001b[0m \u001b[0mcat\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mcat_pol_couples\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m     \u001b[0mc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mtxt\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\" \"\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcat\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlower\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msplit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"#\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0;34m\" \"\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0mcat\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 6\u001b[0;31m     \u001b[0minf\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0minference_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mc\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mth\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      7\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minf\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m>\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      8\u001b[0m       \u001b[0mpredict\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mextend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minf\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m<ipython-input-43-1b038001cc09>\u001b[0m in \u001b[0;36minference_function\u001b[0;34m(c, th)\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0minference_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mc\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mth\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m   \u001b[0mres\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m   \u001b[0mtokens\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtokenize_couple\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      6\u001b[0m   \u001b[0mx\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0marray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtokens\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'input_ids'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0marray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtokens\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'token_type_ids'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0marray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtokens\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'attention_mask'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      7\u001b[0m   \u001b[0mout2\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mner_model\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m<ipython-input-47-bad57c11a90c>\u001b[0m in \u001b[0;36mtokenize_couple\u001b[0;34m(ex)\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mtokenize_couple\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mex\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m   \u001b[0mdct\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mslow_tokenizer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mex\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmax_length\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mmax_len_txt\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpadding\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'max_length'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m   \u001b[0mdct2\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mslow_tokenizer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mex\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0;34m\"[SEP]\"\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0madd_special_tokens\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mmax_length\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mquery_len\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpadding\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'max_length'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m   \u001b[0mdct2\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'token_type_ids'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mquery_len\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m   \u001b[0mdct\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'input_ids'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mextend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdct2\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'input_ids'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mNameError\u001b[0m: name 'max_len_txt' is not defined"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "\n",
        "def jaccardIndex(predict, target):\n",
        "  predict = list(map(str,predict))\n",
        "  target = list(map(str,target))\n",
        "  predict = set(predict)\n",
        "  target = set(target)\n",
        "  \n",
        "  inter = len(predict.intersection(target))\n",
        "  union = len(predict.union(target))\n",
        "  if union==0:\n",
        "    return 1\n",
        "  return inter / union"
      ],
      "metadata": {
        "id": "HgN2Qkd8TBfY"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "\n",
        "def precision(predict, target):\n",
        "  predict = list(map(str,predict))\n",
        "  target = list(map(str,target))\n",
        "  predict = set(predict)\n",
        "  target = set(target)\n",
        "  \n",
        "  inter = len(predict.intersection(target))\n",
        "  num_predict = len(predict)\n",
        "  if num_predict==0:\n",
        "    return 1\n",
        "  return inter / num_predict"
      ],
      "metadata": {
        "id": "C3LB5x96Wjn9"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "\n",
        "def recall(predict, target):\n",
        "  predict = list(map(str,predict))\n",
        "  target = list(map(str,target))\n",
        "  predict = set(predict)\n",
        "  target = set(target)\n",
        "  \n",
        "  inter = len(predict.intersection(target))\n",
        "  num_target = len(target)\n",
        "  if num_target==0:\n",
        "    return 1\n",
        "  return inter / num_target"
      ],
      "metadata": {
        "id": "Yhj7vzWuWykC"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "\n",
        "def f1score(predict, target):\n",
        "  return (2 * precision(predict, target) * recall(predict, target)) / (precision(predict, target) + recall(predict, target))"
      ],
      "metadata": {
        "id": "w48ndgrQAKc5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import zipfile\n",
        "import pandas as pd\n",
        "import xml.etree.ElementTree as ET\n",
        "!pip install tensorflow_text\n",
        "\n",
        "dataset_path = \"/content/drive/MyDrive/HLT/RestaurantTS.zip\"\n",
        "with zipfile.ZipFile(dataset_path,\"r\") as zip_ref:\n",
        "  zip_ref.extractall(\"./dataset\")"
      ],
      "metadata": {
        "id": "by5Roihfl4g0"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "\n",
        "def evaluate_restaurants():\n",
        "  \"\"\"\n",
        "    Mean of Jaccard index over the sentences\n",
        "  \"\"\"\n",
        "\n",
        "  xml_path = \"/content/dataset/ABSA15_Restaurants_Test.xml\"\n",
        "  mytree = ET.parse(xml_path)\n",
        "  reviews = mytree.getroot()\n",
        "  input = []\n",
        "  all_opinions = []\n",
        "  for rew in reviews:\n",
        "    for sent in rew[0]: #cicle on single sentence in review.\n",
        "      txt = sent[0].text\n",
        "      if len(sent)>1:\n",
        "        input.append(txt)\n",
        "        sent_opinions = []\n",
        "        for opinion in sent[1]:\n",
        "          op_dict = opinion.attrib\n",
        "          sent_opinions.append(op_dict)\n",
        "        all_opinions.append(sent_opinions)\n",
        "\n",
        "  JI = 0\n",
        "  PR = 0\n",
        "  RC = 0\n",
        "  F1 = 0\n",
        "  tot = len(input)\n",
        "  i = 0\n",
        "  for sent_txt, opinions_target in zip(input, all_opinions):\n",
        "    predict = get_opinions(sent_txt)\n",
        "    i += 1\n",
        "    JI += jaccardIndex(predict, opinions_target)\n",
        "    PR += precision(predict, opinions_target)\n",
        "    RC += recall(predict, opinions_target)\n",
        "    F1 += f1score(predict, opinions_target)\n",
        "    print(\"Observed={}\".format(i/tot))\n",
        "    print(\"Jaccard={} Precision={} Recall={} F1-score={}\".format(JI/i,PR/i,RC/i,F1/i))\n",
        "\n",
        "  MJI = JI/tot\n",
        "  MPR = PR/tot\n",
        "  MRC = RC/tot\n",
        "  MF1 = F1/tot\n",
        "\n",
        "  return MJI,MPR,MRC,MF1"
      ],
      "metadata": {
        "id": "ivxBt7gRlylP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "j,p,r,f = evaluate_restaurants()\n",
        "print(\"Jaccard:{} Precision:{} Recall:{} F1:{}\".format(j,p,r,f))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "g9gyyU_ufSix",
        "outputId": "c8d93e36-7253-4cba-d0bd-7c22fc2b091a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "0.0\n",
            "J=0.0 P=0.0 R=0.0\n",
            "0.001718213058419244\n",
            "J=0.5 P=0.5 R=0.5\n",
            "0.003436426116838488\n",
            "J=0.5 P=0.5 R=0.6666666666666666\n",
            "0.005154639175257732\n",
            "J=0.5 P=0.5 R=0.75\n",
            "0.006872852233676976\n",
            "J=0.5 P=0.5 R=0.8\n",
            "0.00859106529209622\n",
            "J=0.44047619047619047 P=0.44047619047619047 R=0.8333333333333334\n",
            "0.010309278350515464\n",
            "J=0.401360544217687 P=0.401360544217687 R=0.8571428571428571\n",
            "0.012027491408934709\n",
            "J=0.35119047619047616 P=0.35119047619047616 R=0.75\n",
            "0.013745704467353952\n",
            "J=0.3677248677248677 P=0.3677248677248677 R=0.7777777777777778\n",
            "0.015463917525773196\n",
            "J=0.33095238095238094 P=0.33095238095238094 R=0.7\n",
            "0.01718213058419244\n",
            "J=0.3463203463203463 P=0.3463203463203463 R=0.7272727272727273\n",
            "0.018900343642611683\n",
            "J=0.35912698412698413 P=0.35912698412698413 R=0.75\n",
            "0.020618556701030927\n",
            "J=0.4084249084249084 P=0.4084249084249084 R=0.7692307692307693\n",
            "0.022336769759450172\n",
            "J=0.4149659863945578 P=0.4149659863945578 R=0.7857142857142857\n",
            "0.024054982817869417\n",
            "J=0.3873015873015873 P=0.45396825396825397 R=0.7333333333333333\n",
            "0.02577319587628866\n",
            "J=0.3943452380952381 P=0.4568452380952381 R=0.75\n",
            "0.027491408934707903\n",
            "J=0.3711484593837535 P=0.4299719887955182 R=0.7058823529411765\n",
            "0.029209621993127148\n",
            "J=0.3727513227513228 P=0.4283068783068783 R=0.7222222222222222\n",
            "0.030927835051546393\n",
            "J=0.3531328320802005 P=0.40576441102756894 R=0.6842105263157895\n",
            "0.03264604810996564\n",
            "J=0.3854761904761905 P=0.43547619047619046 R=0.7\n",
            "0.03436426116838488\n",
            "J=0.3909297052154195 P=0.4385487528344671 R=0.7142857142857143\n",
            "0.03608247422680412\n",
            "J=0.37770562770562766 P=0.42366522366522363 R=0.7045454545454546\n",
            "0.037800687285223365\n",
            "J=0.3612836438923395 P=0.4052449965493443 R=0.6739130434782609\n",
            "0.03951890034364261\n",
            "J=0.3628968253968254 P=0.405026455026455 R=0.6875\n",
            "0.041237113402061855\n",
            "J=0.3563809523809523 P=0.39682539682539675 R=0.7\n",
            "0.0429553264604811\n",
            "J=0.3426739926739926 P=0.3815628815628815 R=0.6730769230769231\n",
            "0.044673539518900345\n",
            "J=0.3447971781305114 P=0.385949441504997 R=0.6728395061728395\n",
            "0.04639175257731959\n",
            "J=0.33248299319727886 P=0.3721655328798185 R=0.6488095238095238\n",
            "0.048109965635738834\n",
            "J=0.3253284072249589 P=0.36364258347016964 R=0.6609195402298851\n",
            "0.04982817869415808\n",
            "J=0.3144841269841269 P=0.35152116402116396 R=0.638888888888889\n",
            "0.05154639175257732\n",
            "J=0.320468509984639 P=0.356310803891449 R=0.6505376344086022\n",
            "0.05326460481099656\n",
            "J=0.310453869047619 P=0.3451760912698412 R=0.6302083333333334\n",
            "0.054982817869415807\n",
            "J=0.3111471861471861 P=0.3448172198172198 R=0.6414141414141414\n",
            "0.05670103092783505\n",
            "J=0.30787815126050416 P=0.3420284780578898 R=0.6372549019607844\n",
            "0.058419243986254296\n",
            "J=0.3034772370486656 P=0.3366518402232688 R=0.6476190476190476\n",
            "0.06013745704467354\n",
            "J=0.30893620268620264 P=0.34118928910595575 R=0.6574074074074074\n",
            "0.061855670103092786\n",
            "J=0.30959558459558456 P=0.340976965976966 R=0.6666666666666667\n",
            "0.06357388316151202\n",
            "J=0.308027279737806 P=0.34077581774950194 R=0.662280701754386\n",
            "0.06529209621993128\n",
            "J=0.30012914435991356 P=0.35767900190977114 R=0.6452991452991453\n",
            "0.06701030927835051\n",
            "J=0.30512591575091574 P=0.36123702686202686 R=0.6541666666666667\n",
            "0.06872852233676977\n",
            "J=0.30987894219601536 P=0.3646214896214896 R=0.6626016260162602\n",
            "0.070446735395189\n",
            "J=0.3104373800802372 P=0.3654638351066922 R=0.6626984126984128\n",
            "0.07216494845360824\n",
            "J=0.314845813101627 P=0.36859258312746684 R=0.6705426356589148\n",
            "0.07388316151202749\n",
            "J=0.315265984015984 P=0.3677912365412366 R=0.6780303030303031\n",
            "0.07560137457044673\n",
            "J=0.3156674806674807 P=0.36702550535883866 R=0.6851851851851852\n",
            "0.07731958762886598\n",
            "J=0.3088051441312311 P=0.3590466900249509 R=0.6702898550724639\n",
            "0.07903780068728522\n",
            "J=0.3128731197880134 P=0.36204569662016467 R=0.6773049645390071\n",
            "0.08075601374570447\n",
            "J=0.31677159645909647 P=0.3649197446072446 R=0.6840277777777778\n",
            "0.08247422680412371\n",
            "J=0.3148420173930378 P=0.36200755027285636 R=0.6904761904761906\n",
            "0.08419243986254296\n",
            "J=0.31254517704517704 P=0.3587673992673992 R=0.6966666666666668\n",
            "0.0859106529209622\n",
            "J=0.3260246833776245 P=0.37134058751705806 R=0.7026143790849674\n",
            "0.08762886597938144\n",
            "J=0.31975497792805485 P=0.36419942237249925 R=0.6891025641025641\n",
            "0.08934707903780069\n",
            "J=0.31372186513695943 P=0.3573277351579238 R=0.6761006289308177\n",
            "0.09106529209621993\n",
            "J=0.3140850404739293 P=0.35688339438339434 R=0.6820987654320988\n",
            "0.09278350515463918\n",
            "J=0.31746531246531245 P=0.3594855144855144 R=0.6878787878787879\n",
            "0.09450171821305842\n",
            "J=0.32072486045700327 P=0.3619947017268445 R=0.693452380952381\n",
            "0.09621993127147767\n",
            "J=0.3150981085191611 P=0.3556439174860227 R=0.6812865497076024\n",
            "0.0979381443298969\n",
            "J=0.3096653825102101 P=0.3495121258052292 R=0.6695402298850575\n",
            "0.09965635738831616\n",
            "J=0.3086541048405455 P=0.34782547960514054 R=0.6751412429378532\n",
            "0.1013745704467354\n",
            "J=0.3035098697598697 P=0.3420283882783882 R=0.663888888888889\n",
            "0.10309278350515463\n",
            "J=0.30058347845233085 P=0.33847054584759495 R=0.6693989071038252\n",
            "0.10481099656357389\n",
            "J=0.29976761589664813 P=0.33704360155973057 R=0.674731182795699\n",
            "0.10652920962199312\n",
            "J=0.2981840029459077 P=0.33486830629687764 R=0.6798941798941799\n",
            "0.10824742268041238\n",
            "J=0.29743112789987786 P=0.33354223901098895 R=0.6848958333333334\n",
            "0.10996563573883161\n",
            "J=0.2928552643937259 P=0.34379543533389684 R=0.6743589743589744\n",
            "0.11168384879725086\n",
            "J=0.29599382099382093 P=0.34616217116217113 R=0.6792929292929293\n",
            "0.1134020618556701\n",
            "J=0.30650137590436094 P=0.3559209447269148 R=0.6840796019900498\n",
            "0.11512027491408934\n",
            "J=0.3093469439057674 P=0.3580397543632837 R=0.6887254901960784\n",
            "0.11683848797250859\n",
            "J=0.3084868432694519 P=0.3576816902903859 R=0.6859903381642513\n",
            "0.11855670103092783\n",
            "J=0.30884179312750737 P=0.3573338566195708 R=0.6904761904761905\n",
            "0.12027491408934708\n",
            "J=0.31153416223838754 P=0.35934323892070363 R=0.6948356807511737\n",
            "0.12199312714776632\n",
            "J=0.3072072988739655 P=0.3543523606023605 R=0.6851851851851852\n",
            "0.12371134020618557\n",
            "J=0.30397745329252174 P=0.3504766922575141 R=0.6894977168949772\n",
            "0.1254295532646048\n",
            "J=0.3066264066264066 P=0.35249727749727744 R=0.6936936936936937\n",
            "0.12714776632302405\n",
            "J=0.3047602767602767 P=0.35001953601953595 R=0.6977777777777778\n",
            "0.12886597938144329\n",
            "J=0.30184676434676433 P=0.3468760041128462 R=0.6918859649122807\n",
            "0.13058419243986255\n",
            "J=0.30442018299161155 P=0.34886462743605595 R=0.6958874458874459\n",
            "0.1323024054982818\n",
            "J=0.3037224883378729 P=0.34759713221251676 R=0.6997863247863249\n",
            "0.13402061855670103\n",
            "J=0.3062070138019505 P=0.3495262824376748 R=0.7035864978902954\n",
            "0.13573883161512026\n",
            "J=0.3036294261294261 P=0.3465460927960927 R=0.7010416666666667\n",
            "0.13745704467353953\n",
            "J=0.3029673344488159 P=0.34535416572453603 R=0.7047325102880658\n",
            "0.13917525773195877\n",
            "J=0.2992726108579767 P=0.3411425295571637 R=0.6961382113821138\n",
            "0.140893470790378\n",
            "J=0.29566691675125406 P=0.3370323785986436 R=0.6877510040160643\n",
            "0.14261168384879724\n",
            "J=0.2980994534565963 P=0.3389724693296121 R=0.691468253968254\n",
            "0.14432989690721648\n",
            "J=0.2945924010629893 P=0.33498455792573434 R=0.6833333333333333\n",
            "0.14604810996563575\n",
            "J=0.2911669080273731 P=0.3310893886475281 R=0.6753875968992248\n",
            "0.14776632302405499\n",
            "J=0.2887043352560594 P=0.3283286955700748 R=0.671455938697318\n",
            "0.14948453608247422\n",
            "J=0.2854236041736042 P=0.32459768766586944 R=0.6638257575757577\n",
            "0.15120274914089346\n",
            "J=0.2822165973851367 P=0.32095052263591584 R=0.656367041198502\n",
            "0.15292096219931273\n",
            "J=0.27908085741419075 P=0.3284955168288501 R=0.6490740740740741\n",
            "0.15463917525773196\n",
            "J=0.27601403480524356 P=0.324885675984577 R=0.641941391941392\n",
            "0.1563573883161512\n",
            "J=0.27301388225301265 P=0.3213543099412664 R=0.6349637681159421\n",
            "0.15807560137457044\n",
            "J=0.27007824911050715 P=0.3178988872537259 R=0.628136200716846\n",
            "0.15979381443298968\n",
            "J=0.27252422518379965 P=0.31983613313400544 R=0.6320921985815603\n",
            "0.16151202749140894\n",
            "J=0.27176081228712806 P=0.31880861828230245 R=0.6324561403508773\n",
            "0.16323024054982818\n",
            "J=0.2724021927146927 P=0.3189599173974173 R=0.6362847222222222\n",
            "0.16494845360824742\n",
            "J=0.2695939226867062 P=0.3156716708263099 R=0.6297250859106529\n",
            "0.16666666666666666\n",
            "J=0.2668429642919439 P=0.31245053132808226 R=0.6232993197278912\n",
            "0.16838487972508592\n",
            "J=0.26583108249774917 P=0.31097796703857306 R=0.6271043771043772\n",
            "0.17010309278350516\n",
            "J=0.27317277167277165 P=0.3178681873681873 R=0.6308333333333334\n",
            "0.1718213058419244\n",
            "J=0.27136818077412134 P=0.31562106760126557 R=0.6344884488448845\n",
            "0.17353951890034364\n",
            "J=0.26870770841359076 P=0.3125267434090963 R=0.6282679738562092\n",
            "0.17525773195876287\n",
            "J=0.2693351416652387 P=0.31272874913651605 R=0.6318770226537218\n",
            "0.17697594501718214\n",
            "J=0.2667453806876883 P=0.30972174193328034 R=0.6258012820512822\n",
            "0.17869415807560138\n",
            "J=0.2642049484906627 P=0.3067720110577253 R=0.61984126984127\n",
            "0.18041237113402062\n",
            "J=0.2617124489765999 P=0.303877935481709 R=0.613993710691824\n",
            "0.18213058419243985\n",
            "J=0.2592665382385008 P=0.30103795477627243 R=0.6082554517133957\n",
            "0.18384879725085912\n",
            "J=0.2568659221436999 P=0.2982505663061218 R=0.6026234567901235\n",
            "0.18556701030927836\n",
            "J=0.2590965100139412 P=0.30010147854184543 R=0.6062691131498472\n",
            "0.1872852233676976\n",
            "J=0.25901381446835986 P=0.29964601055510137 R=0.609848484848485\n",
            "0.18900343642611683\n",
            "J=0.25848215848215844 P=0.29894849894849884 R=0.6103603603603605\n",
            "0.19072164948453607\n",
            "J=0.25617428206713916 P=0.296279315922173 R=0.6049107142857144\n",
            "0.19243986254295534\n",
            "J=0.25769992053177887 P=0.29808215383436615 R=0.6061946902654869\n",
            "0.19415807560137457\n",
            "J=0.2642113247376405 P=0.30423932792353836 R=0.6096491228070177\n",
            "0.1958762886597938\n",
            "J=0.26365296539209576 P=0.3033328989850728 R=0.6130434782608697\n",
            "0.19759450171821305\n",
            "J=0.270000784655957 P=0.30933864985589116 R=0.6163793103448277\n",
            "0.19931271477663232\n",
            "J=0.27054208849080635 P=0.30954373262065565 R=0.6196581196581198\n",
            "0.20103092783505155\n",
            "J=0.27248664706291814 P=0.3111577687848874 R=0.6228813559322035\n",
            "0.2027491408934708\n",
            "J=0.2701968433060869 P=0.30854299761862786 R=0.6176470588235295\n",
            "0.20446735395189003\n",
            "J=0.2700285362785362 P=0.3080551393051393 R=0.6208333333333335\n",
            "0.20618556701030927\n",
            "J=0.2677968954828458 P=0.30550922906294803 R=0.6157024793388431\n",
            "0.20790378006872853\n",
            "J=0.27379856027397004 P=0.3112017763657108 R=0.6188524590163935\n",
            "0.20962199312714777\n",
            "J=0.2715725557188971 P=0.3086716806229001 R=0.6138211382113822\n",
            "0.211340206185567\n",
            "J=0.26938245446309955 P=0.30618239287594123 R=0.6088709677419356\n",
            "0.21305841924398625\n",
            "J=0.2698940614940614 P=0.3063996003996004 R=0.6120000000000001\n",
            "0.21477663230240548\n",
            "J=0.26775204513299744 P=0.3039678575392861 R=0.6071428571428572\n",
            "0.21649484536082475\n",
            "J=0.26761226525006043 P=0.30419908175813687 R=0.6062992125984253\n",
            "0.218213058419244\n",
            "J=0.26552154442779435 P=0.30182252643190144 R=0.6015625000000001\n",
            "0.21993127147766323\n",
            "J=0.26346323788184245 P=0.2994828169246774 R=0.5968992248062016\n",
            "0.22164948453608246\n",
            "J=0.2614365975904437 P=0.2971791029483337 R=0.5923076923076924\n",
            "0.22336769759450173\n",
            "J=0.2613492953187609 P=0.2974550894398223 R=0.5916030534351147\n",
            "0.22508591065292097\n",
            "J=0.2593693764148309 P=0.29520164179255093 R=0.5871212121212123\n",
            "0.2268041237113402\n",
            "J=0.2586723635595815 P=0.29423521340814573 R=0.5902255639097745\n",
            "0.22852233676975944\n",
            "J=0.2567419727867488 P=0.2920394282334581 R=0.5858208955223881\n",
            "0.23024054982817868\n",
            "J=0.25484018039573586 P=0.28987617320950654 R=0.5814814814814816\n",
            "0.23195876288659795\n",
            "J=0.2538854731869437 P=0.28866384840649545 R=0.5845588235294119\n",
            "0.23367697594501718\n",
            "J=0.25203229455054266 P=0.286556813016667 R=0.5802919708029198\n",
            "0.23539518900343642\n",
            "J=0.25141370304413774 P=0.2857978374282722 R=0.5809178743961354\n",
            "0.23711340206185566\n",
            "J=0.24960497136756118 P=0.28374173787842855 R=0.57673860911271\n",
            "0.23883161512027493\n",
            "J=0.2506792215720786 P=0.2845721540364397 R=0.5797619047619049\n",
            "0.24054982817869416\n",
            "J=0.24890135475241848 P=0.28255391180923095 R=0.5756501182033098\n",
            "0.2422680412371134\n",
            "J=0.24761801187857518 P=0.28115094998897816 R=0.5733568075117372\n",
            "0.24398625429553264\n",
            "J=0.24625447038954576 P=0.2795733598181151 R=0.572843822843823\n",
            "0.24570446735395188\n",
            "J=0.2514888143451739 P=0.28457632259715593 R=0.5758101851851853\n",
            "0.24742268041237114\n",
            "J=0.2514785466600348 P=0.2843378651999342 R=0.5787356321839082\n",
            "0.24914089347079038\n",
            "J=0.25318074839524 P=0.2869565556209392 R=0.5793378995433791\n",
            "0.2508591065292096\n",
            "J=0.25315911065105473 P=0.2867051504806607 R=0.582199546485261\n",
            "0.25257731958762886\n",
            "J=0.2531377653088179 P=0.2864571427071427 R=0.5850225225225227\n",
            "0.2542955326460481\n",
            "J=0.2547945588302352 P=0.28789031624602096 R=0.5878076062639822\n",
            "0.25601374570446733\n",
            "J=0.254429261771367 P=0.2876377141377141 R=0.5872222222222224\n",
            "0.25773195876288657\n",
            "J=0.2560555580510268 P=0.289044086891769 R=0.5899558498896249\n",
            "0.25945017182130586\n",
            "J=0.26094992937963846 P=0.2937214284253758 R=0.59265350877193\n",
            "0.2611683848797251\n",
            "J=0.26251234814186303 P=0.29506965438337984 R=0.5953159041394337\n",
            "0.26288659793814434\n",
            "J=0.26297222466908043 P=0.29531811983110684 R=0.5979437229437231\n",
            "0.2646048109965636\n",
            "J=0.26127562967121537 P=0.2998644545418739 R=0.5940860215053765\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "opinions = get_opinions(\"I recommend this place to everyone.\")"
      ],
      "metadata": {
        "id": "Vh5d72Aiy7pE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "opinions"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "O5vBH8yz3BwP",
        "outputId": "d48a7a84-9c6b-444a-ad76-6bcda20e4a52"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[{'category': 'RESTAURANT#GENERAL',\n",
              "  'from': '17',\n",
              "  'polarity': 'positive',\n",
              "  'target': 'place',\n",
              "  'to': '22'}]"
            ]
          },
          "metadata": {},
          "execution_count": 86
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "target = [{'target':\"pastas\", 'category':\"FOOD#QUALITY\", 'polarity':\"positive\", 'from':4, 'to':10},\n",
        "{'target':\"risottos\", 'category':\"FOOD#QUALITY\", 'polarity':\"positive\", 'from':31, 'to':39},\n",
        "{'target':\"sepia\", 'category':\"FOOD#QUALITY\", 'polarity':\"positive\", 'from':58, 'to':63},\n",
        "{'target':\"braised rabbit\", 'category':\"FOOD#QUALITY\", 'polarity':\"positive\", 'from':87, 'to':101}]"
      ],
      "metadata": {
        "id": "wHXtDgZEzP5R"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "precision(opinions,target)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6FAhfsWfxt2t",
        "outputId": "e8acd46d-3991-4398-c24d-8971e0f9ba48"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.5"
            ]
          },
          "metadata": {},
          "execution_count": 63
        }
      ]
    }
  ],
  "metadata": {
    "colab": {
      "collapsed_sections": [
        "UMGHVW79BznX",
        "vSlujw-M33Su",
        "RXpxaJE7Vq77",
        "BtJhqPqWZ5CK",
        "zzzzWSjepudC"
      ],
      "name": "hlt_project.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "10a4469cbeb540ff8d961d23b1dc58df": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_95d064bf3eb246d2861fa0a9dfb19ed9",
              "IPY_MODEL_76a8eb371869447488dcc44d8cd315b1",
              "IPY_MODEL_28aa9491c51e4d63994e4c165dff661f"
            ],
            "layout": "IPY_MODEL_792e7c63bdf248c89ff9ef615752867e"
          }
        },
        "95d064bf3eb246d2861fa0a9dfb19ed9": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_00a41a671d3c4d39927d3c88dd2f4a36",
            "placeholder": "​",
            "style": "IPY_MODEL_dec923fa743f4f46b4f4644702c43fc2",
            "value": "Downloading: 100%"
          }
        },
        "76a8eb371869447488dcc44d8cd315b1": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_1cad7d52892e46a6af14a1f6393faafc",
            "max": 536063208,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_c6988e819ce54ad99a847e7e8ada2706",
            "value": 536063208
          }
        },
        "28aa9491c51e4d63994e4c165dff661f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_fdfbfd3a5dae45ce89c91eac4870a009",
            "placeholder": "​",
            "style": "IPY_MODEL_59652c3a3a37432b8c05c52f00af132f",
            "value": " 511M/511M [00:23&lt;00:00, 25.1MB/s]"
          }
        },
        "792e7c63bdf248c89ff9ef615752867e": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "00a41a671d3c4d39927d3c88dd2f4a36": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "dec923fa743f4f46b4f4644702c43fc2": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "1cad7d52892e46a6af14a1f6393faafc": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "c6988e819ce54ad99a847e7e8ada2706": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "fdfbfd3a5dae45ce89c91eac4870a009": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "59652c3a3a37432b8c05c52f00af132f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "77b9ac899d724d0eb4246ba47061eb71": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_b1f31b25e15149b89f6e04827439b1f2",
              "IPY_MODEL_fcec08fd554a4284aafa08eeba9a9fed",
              "IPY_MODEL_9e5b0ae4e386465b892af86c37e21fb7"
            ],
            "layout": "IPY_MODEL_05022614f2f945dfbb79c7ecda2b5cf8"
          }
        },
        "b1f31b25e15149b89f6e04827439b1f2": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_9d5593ff4a6a4653b23eda14d497ccff",
            "placeholder": "​",
            "style": "IPY_MODEL_8da2cbc75ffe4c649838d0dc7a0a1990",
            "value": "Downloading: 100%"
          }
        },
        "fcec08fd554a4284aafa08eeba9a9fed": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_bd36a07085aa40a08adba7844b9e25f6",
            "max": 536063208,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_1397738b8e794d28ad66d22f5fc2a1d1",
            "value": 536063208
          }
        },
        "9e5b0ae4e386465b892af86c37e21fb7": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_26f0dcf3c5624240983563ad63954490",
            "placeholder": "​",
            "style": "IPY_MODEL_dc07f6b87a8d441d8ffe8d7998629e0f",
            "value": " 511M/511M [00:24&lt;00:00, 26.3MB/s]"
          }
        },
        "05022614f2f945dfbb79c7ecda2b5cf8": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "9d5593ff4a6a4653b23eda14d497ccff": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "8da2cbc75ffe4c649838d0dc7a0a1990": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "bd36a07085aa40a08adba7844b9e25f6": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "1397738b8e794d28ad66d22f5fc2a1d1": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "26f0dcf3c5624240983563ad63954490": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "dc07f6b87a8d441d8ffe8d7998629e0f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        }
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}