25_nn_embed_remap.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8816b366-5be4-4748-b769-4dc6f7d1735f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Modify the file 00_setup.py to define input/output file paths on your system\n",
    "# The information in 00_setup.py will be used across notebooks\n",
    "from importlib.machinery import SourceFileLoader\n",
    "setup = SourceFileLoader(\"setup\", \"./00_setup.py\").load_module()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a945de0c-f042-46b2-9d06-ab064af6c05b",
   "metadata": {},
   "source": [
    "## Neural Network With NAICS Embedding Layer, Modified\n",
    "Add NAICS embedding to model.  Modify the training data to set some NAICS to the missing value (0) in the input"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d47a7ec9-7d7b-47df-a902-b2f1e54f1135",
   "metadata": {},
   "source": [
    "*This script takes about 2 hours on my MacBook Air*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "fb42f9b4-99a8-4753-aacf-8722469be4c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "bf51170a-4487-4d10-b63d-c88f41b529f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "import importlib, pickle\n",
    "from sklearn import model_selection\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "09dd973b-60a0-434b-9193-8c5fcec27dd5",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import os\n",
    "import re\n",
    "\n",
    "import keras\n",
    "from tensorflow.keras import layers, optimizers, losses, metrics, Model\n",
    "from sklearn import preprocessing, feature_extraction, model_selection\n",
    "from IPython.display import display, HTML\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "22b466e7-a419-45aa-83f7-5dafc7a3998b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import Dense, Dropout, Input, Embedding, Concatenate, Reshape\n",
    "import tensorflow.keras.metrics as km"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "9f1e4f2e-0e45-4e28-8e9c-212e0bb51199",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sba_gnn.sba_gnn import sg_plot "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3cbc00d2-bc83-42d9-83eb-65fcf306dccd",
   "metadata": {},
   "source": [
    "## Input Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "5f0183dc-7ee1-49eb-8723-a9f4cea14b88",
   "metadata": {},
   "outputs": [],
   "source": [
    "business_data = pd.read_parquet(Path(setup.temp_path).joinpath('20_DATA_combined_scaled_all.parquet'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "feaedcfe-7722-41c1-be4c-4921f0d11db9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<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>target</th>\n",
       "      <th>dset_naics_holdout</th>\n",
       "      <th>mhier_NAICS</th>\n",
       "      <th>menc_NAICS</th>\n",
       "      <th>cenc_NAICS</th>\n",
       "      <th>NS___Accommodation and Food Services</th>\n",
       "      <th>NS___Administrative and Support and Waste Management and Remediation Services</th>\n",
       "      <th>NS___Construction</th>\n",
       "      <th>NS___Health Care and Social Assistance</th>\n",
       "      <th>NS___Manufacturing</th>\n",
       "      <th>...</th>\n",
       "      <th>new_business</th>\n",
       "      <th>urban_flag</th>\n",
       "      <th>franchise_flag</th>\n",
       "      <th>missingindicator_LowDoc</th>\n",
       "      <th>missingindicator_new_business</th>\n",
       "      <th>missingindicator_urban_flag</th>\n",
       "      <th>NAICS</th>\n",
       "      <th>NAICS_alt</th>\n",
       "      <th>NAICS_alt3</th>\n",
       "      <th>NAICS_alt4</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.203011</td>\n",
       "      <td>0.094541</td>\n",
       "      <td>0.203906</td>\n",
       "      <td>0.205152</td>\n",
       "      <td>0.007265</td>\n",
       "      <td>0.097958</td>\n",
       "      <td>0.047133</td>\n",
       "      <td>0.095606</td>\n",
       "      <td>0.079398</td>\n",
       "      <td>0.096843</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.444129</td>\n",
       "      <td>0.705279</td>\n",
       "      <td>-0.899352</td>\n",
       "      <td>-0.985316</td>\n",
       "      <td>-0.997529</td>\n",
       "      <td>-0.602849</td>\n",
       "      <td>701.045842</td>\n",
       "      <td>767.921227</td>\n",
       "      <td>769.528095</td>\n",
       "      <td>769.233577</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.402241</td>\n",
       "      <td>0.292580</td>\n",
       "      <td>0.095690</td>\n",
       "      <td>0.091362</td>\n",
       "      <td>0.010443</td>\n",
       "      <td>0.297258</td>\n",
       "      <td>0.211922</td>\n",
       "      <td>0.294051</td>\n",
       "      <td>0.270359</td>\n",
       "      <td>0.295744</td>\n",
       "      <td>...</td>\n",
       "      <td>0.895963</td>\n",
       "      <td>0.708930</td>\n",
       "      <td>0.437226</td>\n",
       "      <td>0.170743</td>\n",
       "      <td>0.070251</td>\n",
       "      <td>0.797856</td>\n",
       "      <td>383.284385</td>\n",
       "      <td>326.131581</td>\n",
       "      <td>324.361889</td>\n",
       "      <td>326.453771</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.012987</td>\n",
       "      <td>0.020495</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.126003</td>\n",
       "      <td>0.137475</td>\n",
       "      <td>0.000718</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>436.000000</td>\n",
       "      <td>651.000000</td>\n",
       "      <td>644.000000</td>\n",
       "      <td>638.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.205592</td>\n",
       "      <td>0.204655</td>\n",
       "      <td>0.003056</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>756.000000</td>\n",
       "      <td>844.000000</td>\n",
       "      <td>825.000000</td>\n",
       "      <td>825.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.264286</td>\n",
       "      <td>0.254574</td>\n",
       "      <td>0.009309</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>-1.000000</td>\n",
       "      <td>1031.000000</td>\n",
       "      <td>1034.000000</td>\n",
       "      <td>1034.000000</td>\n",
       "      <td>1034.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.572148</td>\n",
       "      <td>0.551324</td>\n",
       "      <td>0.044873</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1166.000000</td>\n",
       "      <td>1166.000000</td>\n",
       "      <td>1166.000000</td>\n",
       "      <td>1166.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 29 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              target  dset_naics_holdout    mhier_NAICS     menc_NAICS  \\\n",
       "count  688081.000000       688081.000000  688081.000000  688081.000000   \n",
       "mean        0.203011            0.094541       0.203906       0.205152   \n",
       "std         0.402241            0.292580       0.095690       0.091362   \n",
       "min         0.000000            0.000000       0.012987       0.020495   \n",
       "25%         0.000000            0.000000       0.126003       0.137475   \n",
       "50%         0.000000            0.000000       0.205592       0.204655   \n",
       "75%         0.000000            0.000000       0.264286       0.254574   \n",
       "max         1.000000            1.000000       0.572148       0.551324   \n",
       "\n",
       "          cenc_NAICS  NS___Accommodation and Food Services  \\\n",
       "count  688081.000000                         688081.000000   \n",
       "mean        0.007265                              0.097958   \n",
       "std         0.010443                              0.297258   \n",
       "min         0.000000                              0.000000   \n",
       "25%         0.000718                              0.000000   \n",
       "50%         0.003056                              0.000000   \n",
       "75%         0.009309                              0.000000   \n",
       "max         0.044873                              1.000000   \n",
       "\n",
       "       NS___Administrative and Support and Waste Management and Remediation Services  \\\n",
       "count                                      688081.000000                               \n",
       "mean                                            0.047133                               \n",
       "std                                             0.211922                               \n",
       "min                                             0.000000                               \n",
       "25%                                             0.000000                               \n",
       "50%                                             0.000000                               \n",
       "75%                                             0.000000                               \n",
       "max                                             1.000000                               \n",
       "\n",
       "       NS___Construction  NS___Health Care and Social Assistance  \\\n",
       "count      688081.000000                           688081.000000   \n",
       "mean            0.095606                                0.079398   \n",
       "std             0.294051                                0.270359   \n",
       "min             0.000000                                0.000000   \n",
       "25%             0.000000                                0.000000   \n",
       "50%             0.000000                                0.000000   \n",
       "75%             0.000000                                0.000000   \n",
       "max             1.000000                                1.000000   \n",
       "\n",
       "       NS___Manufacturing  ...   new_business     urban_flag  franchise_flag  \\\n",
       "count       688081.000000  ...  688081.000000  688081.000000   688081.000000   \n",
       "mean             0.096843  ...      -0.444129       0.705279       -0.899352   \n",
       "std              0.295744  ...       0.895963       0.708930        0.437226   \n",
       "min              0.000000  ...      -1.000000      -1.000000       -1.000000   \n",
       "25%              0.000000  ...      -1.000000       1.000000       -1.000000   \n",
       "50%              0.000000  ...      -1.000000       1.000000       -1.000000   \n",
       "75%              0.000000  ...       1.000000       1.000000       -1.000000   \n",
       "max              1.000000  ...       1.000000       1.000000        1.000000   \n",
       "\n",
       "       missingindicator_LowDoc  missingindicator_new_business  \\\n",
       "count            688081.000000                  688081.000000   \n",
       "mean                 -0.985316                      -0.997529   \n",
       "std                   0.170743                       0.070251   \n",
       "min                  -1.000000                      -1.000000   \n",
       "25%                  -1.000000                      -1.000000   \n",
       "50%                  -1.000000                      -1.000000   \n",
       "75%                  -1.000000                      -1.000000   \n",
       "max                   1.000000                       1.000000   \n",
       "\n",
       "       missingindicator_urban_flag          NAICS      NAICS_alt  \\\n",
       "count                688081.000000  688081.000000  688081.000000   \n",
       "mean                     -0.602849     701.045842     767.921227   \n",
       "std                       0.797856     383.284385     326.131581   \n",
       "min                      -1.000000       0.000000       1.000000   \n",
       "25%                      -1.000000     436.000000     651.000000   \n",
       "50%                      -1.000000     756.000000     844.000000   \n",
       "75%                      -1.000000    1031.000000    1034.000000   \n",
       "max                       1.000000    1166.000000    1166.000000   \n",
       "\n",
       "          NAICS_alt3     NAICS_alt4  \n",
       "count  688081.000000  688081.000000  \n",
       "mean      769.528095     769.233577  \n",
       "std       324.361889     326.453771  \n",
       "min         0.000000       0.000000  \n",
       "25%       644.000000     638.000000  \n",
       "50%       825.000000     825.000000  \n",
       "75%      1034.000000    1034.000000  \n",
       "max      1166.000000    1166.000000  \n",
       "\n",
       "[8 rows x 29 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "business_data.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "6f0ce28e-a80c-415b-b068-5c4d49370da8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1166"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "with open(Path(setup.temp_path).joinpath('20_DATA_naics_max_levels.pkl'), 'rb') as fin:\n",
    "    naics_max_levels= pickle.load(fin)\n",
    "naics_max_levels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "555f4031-e66d-4476-a37e-50e2d382ecf4",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(Path(setup.temp_path).joinpath('20_DATA_features.pkl'), 'rb') as fin:\n",
    "    imputer_features = pickle.load(fin)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "3a916e26-2419-4bd0-8fd4-58a959dc4bac",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['NoEmp', 'CreateJob', 'LowDoc', 'DisbursementGross', 'new_business', 'urban_flag', 'franchise_flag', 'missingindicator_LowDoc', 'missingindicator_new_business', 'missingindicator_urban_flag']\n"
     ]
    }
   ],
   "source": [
    "numeric_features = [f for f in imputer_features if 'NAICS' not in f]\n",
    "print(numeric_features)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d94006a0-f010-4874-a7f1-0fd80c1e9a35",
   "metadata": {},
   "source": [
    "##### Datasets for train, validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "a02a909d-834a-46a8-9cdd-ccfa36b07d69",
   "metadata": {},
   "outputs": [],
   "source": [
    "X = business_data[['dset', 'LoanNr_ChkDgt'] + numeric_features].set_index('LoanNr_ChkDgt').sort_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "42dc5c92-f5a2-457c-be3a-60cb0f6c2ef8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training X: (436120, 10), y:(436120,)\n"
     ]
    }
   ],
   "source": [
    "X_train = X[X['dset'] == 'train'].drop(columns='dset')\n",
    "y_train = business_data[business_data['dset'] == 'train'].set_index('LoanNr_ChkDgt').sort_index()['target']\n",
    "print(f'training X: {X_train.shape}, y:{y_train.shape}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "038b0e4f-a84a-4c8b-863e-4c594c1c9f32",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "val X: (93454, 10), y:(93454,)\n"
     ]
    }
   ],
   "source": [
    "X_val = X[X['dset'] == 'val'].drop(columns='dset')\n",
    "y_val = business_data[business_data['dset'] == 'val'].set_index('LoanNr_ChkDgt').sort_index()['target']\n",
    "print(f'val X: {X_val.shape}, y:{y_val.shape}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "bcb86909-e8a7-4f22-b089-6ca61dd86c73",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.2046546821975603\n"
     ]
    }
   ],
   "source": [
    "base_thresh = y_train.mean()\n",
    "print(base_thresh)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "1b8ce4c9-6907-4b70-982a-b92a52b3159d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Copy frame as I will remap some NAICS\n",
    "X_naics = business_data[['dset', 'LoanNr_ChkDgt', 'NAICS']].copy() \\\n",
    "    .set_index('LoanNr_ChkDgt').sort_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "a6f129e0-a4d0-4d93-9886-153c9803d403",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_naics_train = X_naics[X_naics['dset'] == 'train'].drop(columns='dset')\n",
    "X_naics_val = X_naics[X_naics['dset'] == 'val'].drop(columns='dset')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7543db04-1ded-4e75-8a47-b59f04178064",
   "metadata": {},
   "source": [
    "##### Remap 10% of train NAICS to 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "16c8c0f5-9bae-4e37-97d5-b7d2c6c9501c",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_remap_ind, _ = train_test_split(X_naics_train.index, train_size = 0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "70584b90-91fb-4ea7-8444-44cd628bf933",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train len: 436120, remapping 43612 to 0\n"
     ]
    }
   ],
   "source": [
    "print(f'train len: {len(X_naics_train)}, remapping {len(train_remap_ind)} to 0' )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "bcad25e7-91ad-4fb4-8fbe-cbdba816f5a4",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_naics_train.loc[train_remap_ind, 'NAICS'] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "21ca5700-4e84-4f71-b0dd-f151141c0b5e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    436120.000000\n",
       "mean        696.705100\n",
       "std         385.981606\n",
       "min           0.000000\n",
       "25%         413.000000\n",
       "50%         756.000000\n",
       "75%        1031.000000\n",
       "max        1166.000000\n",
       "Name: NAICS, dtype: float64"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_naics_train['NAICS'].describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a5c68623-4599-4472-a13a-9400d101633e",
   "metadata": {},
   "source": [
    "## Function to create model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "9f4deffd-2b85-43d0-9cd1-1d9cf707b9e7",
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_model(n_feat = len(numeric_features), naics_max_levels = naics_max_levels,\n",
    "                 naics_emd_dim = setup.nn_naics_embed_size,\n",
    "                 hidden_size = setup.nn_layer_sizes,\n",
    "                 activation='tanh', lr=setup.nn_learning_rate,\n",
    "                 opt_func = setup.nn_optimizer, dropout = setup.nn_dropout,\n",
    "                embed_layer_name = 'naics_reshape'):\n",
    "    \n",
    "    n_layers = len(hidden_size)\n",
    "    \n",
    "    features_in = Input(shape=(n_feat,))\n",
    "    naics_in = Input(shape=(1,))\n",
    "    embed_layer = Embedding(naics_max_levels, output_dim=naics_emd_dim, input_length=1)(naics_in)\n",
    "    embed_layer = Reshape(target_shape=(naics_emd_dim,), name=embed_layer_name)(embed_layer)\n",
    "    x = Concatenate(axis=-1)([features_in, embed_layer])\n",
    "    \n",
    "    for i in range(n_layers):\n",
    "        x = Dense(hidden_size[i],activation=activation,\n",
    "                  name=f'layer_{i:02d}')(x)\n",
    "        x = Dropout(dropout, name=f'dropout_{i:02d}')(x)\n",
    "    output = Dense(1, activation='sigmoid')(x)\n",
    "    \n",
    "    model = Model(inputs=[features_in, naics_in], outputs=output)\n",
    "    \n",
    "    # Compile model\n",
    "    optimizer = opt_func(learning_rate=lr)\n",
    "    model.compile(loss='binary_crossentropy', optimizer=optimizer, \n",
    "                  metrics=[km.AUC(curve='PR'), km.AUC(curve='ROC')])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3db4ea1c-77b2-4721-b106-1e66fc502139",
   "metadata": {},
   "source": [
    "## Create, fit model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "4d48ddfa-580c-4321-9917-0bd176b67d87",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-08 06:28:43.310687: I metal_plugin/src/device/metal_device.cc:1154] Metal device set to: Apple M1\n",
      "2024-04-08 06:28:43.310735: I metal_plugin/src/device/metal_device.cc:296] systemMemory: 16.00 GB\n",
      "2024-04-08 06:28:43.310743: I metal_plugin/src/device/metal_device.cc:313] maxCacheSize: 5.33 GB\n",
      "2024-04-08 06:28:43.311303: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:303] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.\n",
      "2024-04-08 06:28:43.311617: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:269] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)\n"
     ]
    }
   ],
   "source": [
    "this_model = create_model()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "47036621-ef13-4957-997b-f8ef86016f7f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                Output Shape                 Param #   Connected to                  \n",
      "==================================================================================================\n",
      " input_2 (InputLayer)        [(None, 1)]                  0         []                            \n",
      "                                                                                                  \n",
      " embedding (Embedding)       (None, 1, 8)                 9328      ['input_2[0][0]']             \n",
      "                                                                                                  \n",
      " input_1 (InputLayer)        [(None, 10)]                 0         []                            \n",
      "                                                                                                  \n",
      " naics_reshape (Reshape)     (None, 8)                    0         ['embedding[0][0]']           \n",
      "                                                                                                  \n",
      " concatenate (Concatenate)   (None, 18)                   0         ['input_1[0][0]',             \n",
      "                                                                     'naics_reshape[0][0]']       \n",
      "                                                                                                  \n",
      " layer_00 (Dense)            (None, 128)                  2432      ['concatenate[0][0]']         \n",
      "                                                                                                  \n",
      " dropout_00 (Dropout)        (None, 128)                  0         ['layer_00[0][0]']            \n",
      "                                                                                                  \n",
      " layer_01 (Dense)            (None, 64)                   8256      ['dropout_00[0][0]']          \n",
      "                                                                                                  \n",
      " dropout_01 (Dropout)        (None, 64)                   0         ['layer_01[0][0]']            \n",
      "                                                                                                  \n",
      " dense (Dense)               (None, 1)                    65        ['dropout_01[0][0]']          \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 20081 (78.44 KB)\n",
      "Trainable params: 20081 (78.44 KB)\n",
      "Non-trainable params: 0 (0.00 Byte)\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "this_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "31772016-3643-468b-8d65-c456168cad6e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-08 06:28:44.220157: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "13629/13629 [==============================] - ETA: 0s - loss: 0.4662 - auc: 0.3349 - auc_1: 0.6968"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-08 06:33:35.356651: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "13629/13629 [==============================] - 340s 25ms/step - loss: 0.4662 - auc: 0.3349 - auc_1: 0.6968 - val_loss: 0.4601 - val_auc: 0.3683 - val_auc_1: 0.7176\n",
      "Epoch 2/20\n",
      "13629/13629 [==============================] - 400s 29ms/step - loss: 0.4584 - auc: 0.3568 - auc_1: 0.7135 - val_loss: 0.4584 - val_auc: 0.3749 - val_auc_1: 0.7213\n",
      "Epoch 3/20\n",
      "13629/13629 [==============================] - 190s 14ms/step - loss: 0.4560 - auc: 0.3651 - auc_1: 0.7186 - val_loss: 0.4574 - val_auc: 0.3794 - val_auc_1: 0.7227\n",
      "Epoch 4/20\n",
      "13629/13629 [==============================] - 189s 14ms/step - loss: 0.4549 - auc: 0.3684 - auc_1: 0.7209 - val_loss: 0.4573 - val_auc: 0.3794 - val_auc_1: 0.7237\n",
      "Epoch 5/20\n",
      "13629/13629 [==============================] - 190s 14ms/step - loss: 0.4540 - auc: 0.3711 - auc_1: 0.7226 - val_loss: 0.4579 - val_auc: 0.3793 - val_auc_1: 0.7241\n",
      "Epoch 6/20\n",
      "13629/13629 [==============================] - 191s 14ms/step - loss: 0.4535 - auc: 0.3727 - auc_1: 0.7238 - val_loss: 0.4564 - val_auc: 0.3820 - val_auc_1: 0.7253\n",
      "Epoch 7/20\n",
      "13629/13629 [==============================] - 190s 14ms/step - loss: 0.4530 - auc: 0.3744 - auc_1: 0.7248 - val_loss: 0.4556 - val_auc: 0.3841 - val_auc_1: 0.7264\n",
      "Epoch 8/20\n",
      "13629/13629 [==============================] - 190s 14ms/step - loss: 0.4526 - auc: 0.3762 - auc_1: 0.7256 - val_loss: 0.4572 - val_auc: 0.3836 - val_auc_1: 0.7264\n",
      "Epoch 9/20\n",
      "13629/13629 [==============================] - 2018s 148ms/step - loss: 0.4521 - auc: 0.3775 - auc_1: 0.7264 - val_loss: 0.4566 - val_auc: 0.3827 - val_auc_1: 0.7264\n",
      "Epoch 10/20\n",
      "13629/13629 [==============================] - 1084s 80ms/step - loss: 0.4519 - auc: 0.3778 - auc_1: 0.7271 - val_loss: 0.4559 - val_auc: 0.3837 - val_auc_1: 0.7262\n",
      "Epoch 11/20\n",
      "13629/13629 [==============================] - 2014s 148ms/step - loss: 0.4516 - auc: 0.3791 - auc_1: 0.7275 - val_loss: 0.4557 - val_auc: 0.3848 - val_auc_1: 0.7265\n",
      "Epoch 12/20\n",
      "13629/13629 [==============================] - 1937s 142ms/step - loss: 0.4515 - auc: 0.3795 - auc_1: 0.7277 - val_loss: 0.4559 - val_auc: 0.3845 - val_auc_1: 0.7265\n",
      "Epoch 13/20\n",
      "13629/13629 [==============================] - 1084s 80ms/step - loss: 0.4512 - auc: 0.3800 - auc_1: 0.7283 - val_loss: 0.4570 - val_auc: 0.3858 - val_auc_1: 0.7260\n",
      "Epoch 14/20\n",
      "13629/13629 [==============================] - 2590s 190ms/step - loss: 0.4511 - auc: 0.3809 - auc_1: 0.7284 - val_loss: 0.4569 - val_auc: 0.3865 - val_auc_1: 0.7270\n",
      "Epoch 15/20\n",
      "13629/13629 [==============================] - 1083s 79ms/step - loss: 0.4507 - auc: 0.3824 - auc_1: 0.7291 - val_loss: 0.4557 - val_auc: 0.3860 - val_auc_1: 0.7265\n",
      "Epoch 16/20\n",
      "13629/13629 [==============================] - 461s 34ms/step - loss: 0.4507 - auc: 0.3824 - auc_1: 0.7291 - val_loss: 0.4559 - val_auc: 0.3863 - val_auc_1: 0.7267\n",
      "Epoch 17/20\n",
      "13629/13629 [==============================] - 183s 13ms/step - loss: 0.4505 - auc: 0.3831 - auc_1: 0.7298 - val_loss: 0.4562 - val_auc: 0.3864 - val_auc_1: 0.7264\n",
      "Epoch 18/20\n",
      "13629/13629 [==============================] - 1083s 79ms/step - loss: 0.4502 - auc: 0.3842 - auc_1: 0.7300 - val_loss: 0.4561 - val_auc: 0.3855 - val_auc_1: 0.7266\n",
      "Epoch 19/20\n",
      "13629/13629 [==============================] - 818s 60ms/step - loss: 0.4500 - auc: 0.3842 - auc_1: 0.7305 - val_loss: 0.4553 - val_auc: 0.3862 - val_auc_1: 0.7271\n",
      "Epoch 20/20\n",
      "13629/13629 [==============================] - 183s 13ms/step - loss: 0.4498 - auc: 0.3850 - auc_1: 0.7309 - val_loss: 0.4566 - val_auc: 0.3845 - val_auc_1: 0.7260\n"
     ]
    }
   ],
   "source": [
    "this_history = this_model.fit([X_train, X_naics_train], y_train,\n",
    "                              validation_data=([X_val, X_naics_val], y_val),\n",
    "                              batch_size=setup.nn_batch_size,\n",
    "                              epochs=setup.nn_epochs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "460c4b43-c673-4ee1-b070-d2d596ac4561",
   "metadata": {},
   "outputs": [],
   "source": [
    "this_history_df = pd.DataFrame(this_history.history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "c85bd09a-2fa9-4255-958b-4d48b17085ff",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Rename columns\n",
    "try:\n",
    "    this_history_df.columns = ['_'.join(c.split('_')[0:-1])  \\\n",
    "                               if re.search(r'_\\d+$', c) else c for c in this_history_df.columns]\n",
    "except:\n",
    "    pass\n",
    "try:\n",
    "    cur_col = list(this_history_df.columns)\n",
    "    this_history_df.columns = [cur_col[0]] + \\\n",
    "        [f'{cur_col[i]}_roc'  if (cur_col[i] == cur_col[i-1]) and 'auc'in cur_col[i] \\\n",
    "         else cur_col[i] for i in range(1, len(cur_col))]\n",
    "except:\n",
    "    pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "8ea25384-9fc0-4916-a577-4bd9bc9b203f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['loss', 'auc', 'auc_roc', 'val_loss', 'val_auc', 'val_auc_roc'], dtype='object')"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "this_history_df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "ad4783f2-8d0b-4850-a45a-3a4be8e63318",
   "metadata": {},
   "outputs": [],
   "source": [
    "this_history_df.to_csv(Path(setup.temp_path).joinpath('25_REPORT_fit_history.csv'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "73e89efd-75a9-4b27-81a8-13e12d5482be",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "this_history_df[['loss', 'val_loss']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "2342a31f-4891-4834-851a-b90b8b729c3c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "this_history_df[['auc', 'val_auc']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "cee80637-cabf-4974-b476-08c5fba165ff",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "this_history_df[['auc_roc', 'val_auc_roc']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "e184b0ca-1d31-4798-88a0-4e590b67062e",
   "metadata": {},
   "outputs": [],
   "source": [
    "this_model.save(Path(setup.temp_path).joinpath('25_DATA_model.keras'),save_format='tf')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d326c9ea-4759-4418-a64d-56c151049c25",
   "metadata": {},
   "source": [
    "## Predictions on all data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "cd6910ee-5ffe-4def-9aa8-7dba76446b08",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "   98/21503 [..............................] - ETA: 33s"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-08 11:02:23.153419: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "21503/21503 [==============================] - 84s 4ms/step\n"
     ]
    }
   ],
   "source": [
    "all_predictions = this_model.predict([X.drop(columns='dset'), \n",
    "                                      X_naics.drop(columns='dset')])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "98a2cce1-b7d3-4ebc-8869-80705297e145",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_predictions_df = pd.DataFrame(all_predictions, index=X.index) \\\n",
    "    .set_axis(['predict_prob'], axis=1) \\\n",
    "    .reset_index() \\\n",
    "    .merge(business_data[['target', 'LoanNr_ChkDgt', 'dset', 'dset_naics_holdout', 'NAICS']], \n",
    "           on='LoanNr_ChkDgt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "3fc7da9c-6bd2-4d1b-b77f-b62679825d5d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<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>predict_prob</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>predict_prob</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.327163</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>target</th>\n",
       "      <td>0.327163</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              predict_prob    target\n",
       "predict_prob      1.000000  0.327163\n",
       "target            0.327163  1.000000"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_predictions_df[['predict_prob', 'target']].corr(method='spearman')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e79afc4e-cb67-4064-9b19-dc52500baef2",
   "metadata": {},
   "source": [
    "##### Threshold Tune & Binary Predictions\n",
    "Using training probability predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "ccd0bc23-fbc0-40d0-8d8b-f2302405a875",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_pred_train = all_predictions_df[all_predictions_df['dset'] == 'train']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "df36d9e9-f35e-46b3-a5be-08081f998319",
   "metadata": {},
   "outputs": [],
   "source": [
    "thresh_tune_data = sg_plot.get_f1_frame(all_pred_train['target'], \n",
    "                                        all_pred_train['predict_prob'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "de634156-caf6-4a52-b5d7-44b8e986aa1d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<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>thresh</th>\n",
       "      <th>f1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>0.22</td>\n",
       "      <td>0.458800</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>0.24</td>\n",
       "      <td>0.458128</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.20</td>\n",
       "      <td>0.456673</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    thresh        f1\n",
       "11    0.22  0.458800\n",
       "12    0.24  0.458128\n",
       "10    0.20  0.456673"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "thresh_tune_data.sort_values('f1', ascending=False, inplace=True)\n",
    "thresh_tune_data.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "c746b880-d781-4bc1-9df2-89211127c22a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.22"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_thresh = thresh_tune_data['thresh'].iloc[0]\n",
    "best_thresh"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3c0ac5e2-be0d-42f0-960d-1f81b4a5d4e2",
   "metadata": {},
   "source": [
    "##### Append binary predictions to probability predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "641da72e-1bca-4fe7-96c4-6c029b521bd7",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_predictions_df['predict_bin'] = sg_plot.get_binary_predictions(all_predictions_df['predict_prob'], best_thresh)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "9f6ecdf4-ac01-4c39-8315-e8583daa337e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "predict_bin\n",
       "0    0.569641\n",
       "1    0.430359\n",
       "Name: proportion, dtype: float64"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_predictions_df['predict_bin'].value_counts(normalize=True, dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "cbfed830-7ff8-40f6-a691-6b894b26d6bc",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_predictions_df.to_parquet(Path(setup.temp_path).joinpath('25_DATA_predictions.parquet'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb0ecbd2-abff-46d8-8c70-f7bccd363b69",
   "metadata": {},
   "source": [
    "## Metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "be4ceefe-c7a8-45bd-8573-2daefe99a1c2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<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>dset</th>\n",
       "      <th>accuracy_score</th>\n",
       "      <th>f1_score</th>\n",
       "      <th>precision_score</th>\n",
       "      <th>recall_score</th>\n",
       "      <th>average_precision_score</th>\n",
       "      <th>roc_auc_score</th>\n",
       "      <th>dset_naics_holdout</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>test</td>\n",
       "      <td>0.647176</td>\n",
       "      <td>0.438374</td>\n",
       "      <td>0.318474</td>\n",
       "      <td>0.703067</td>\n",
       "      <td>0.367152</td>\n",
       "      <td>0.728074</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>train</td>\n",
       "      <td>0.656723</td>\n",
       "      <td>0.458800</td>\n",
       "      <td>0.338674</td>\n",
       "      <td>0.710982</td>\n",
       "      <td>0.397626</td>\n",
       "      <td>0.738958</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>val</td>\n",
       "      <td>0.649742</td>\n",
       "      <td>0.450853</td>\n",
       "      <td>0.334112</td>\n",
       "      <td>0.692986</td>\n",
       "      <td>0.384538</td>\n",
       "      <td>0.726052</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>test</td>\n",
       "      <td>0.650869</td>\n",
       "      <td>0.451593</td>\n",
       "      <td>0.333573</td>\n",
       "      <td>0.698850</td>\n",
       "      <td>0.381034</td>\n",
       "      <td>0.727966</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>test</td>\n",
       "      <td>0.641871</td>\n",
       "      <td>0.418752</td>\n",
       "      <td>0.296957</td>\n",
       "      <td>0.709923</td>\n",
       "      <td>0.343123</td>\n",
       "      <td>0.727708</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    dset  accuracy_score  f1_score  precision_score  recall_score  \\\n",
       "0   test        0.647176  0.438374         0.318474      0.703067   \n",
       "1  train        0.656723  0.458800         0.338674      0.710982   \n",
       "2    val        0.649742  0.450853         0.334112      0.692986   \n",
       "0   test        0.650869  0.451593         0.333573      0.698850   \n",
       "1   test        0.641871  0.418752         0.296957      0.709923   \n",
       "\n",
       "   average_precision_score  roc_auc_score  dset_naics_holdout  \n",
       "0                 0.367152       0.728074                 NaN  \n",
       "1                 0.397626       0.738958                 NaN  \n",
       "2                 0.384538       0.726052                 NaN  \n",
       "0                 0.381034       0.727966                 0.0  \n",
       "1                 0.343123       0.727708                 1.0  "
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_dset_df = all_predictions_df.groupby('dset') \\\n",
    "    .apply(lambda x: sg_plot.dset_metrics(x.target, x.predict_bin, x.predict_prob)) \\\n",
    "    .reset_index()\n",
    "metrics_test_df = all_predictions_df[all_predictions_df['dset'] == 'test'] \\\n",
    "    .groupby(['dset', 'dset_naics_holdout']) \\\n",
    "    .apply(lambda x: sg_plot.dset_metrics(x.target, x.predict_bin, x.predict_prob)) \\\n",
    "    .reset_index()\n",
    "metrics_df = pd.concat([metrics_dset_df, metrics_test_df])\n",
    "metrics_df.to_csv(Path(setup.temp_path).joinpath('25_REPORT_metrics.csv'), index=True)\n",
    "metrics_df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2bf6b2f6-992d-4c4b-88cb-02eb418ae751",
   "metadata": {},
   "source": [
    "## Embeddings\n",
    "Get the NAICS embeddings for all codes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "802d63c7-6106-4d32-8c2b-bfabf7db8b06",
   "metadata": {},
   "outputs": [],
   "source": [
    "#this_model = keras.models.load_model(Path(setup.temp_path).joinpath('25_DATA_model.keras'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "b647025a-000d-4d60-bb42-e623160d0f4e",
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_layer_name = 'naics_reshape'\n",
    "#embed_layer_name = 'reshape_4'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "25b09f02-4c5b-42ad-99a8-0e0c11f23be9",
   "metadata": {},
   "outputs": [],
   "source": [
    "unique_naics_x = business_data.drop_duplicates('NAICS_orig') \\\n",
    "    [['LoanNr_ChkDgt'] + numeric_features].set_index('LoanNr_ChkDgt').sort_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "a4bd9290-cf45-437d-9986-a29c1dcc8547",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1311, 10)"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unique_naics_x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "1cd7de97-c893-4697-9545-b31dc12c3c7e",
   "metadata": {},
   "outputs": [],
   "source": [
    "unique_naics_x_naics = business_data.drop_duplicates('NAICS_orig')[['LoanNr_ChkDgt', 'NAICS']] \\\n",
    "    .set_index('LoanNr_ChkDgt').sort_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "f8bf0603-5235-4344-a73d-a5d632befd0d",
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_model = Model(inputs=this_model.inputs, outputs=this_model.get_layer(embed_layer_name).output) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "0ee68a3c-a630-4fde-b765-f441d2bb3895",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "41/41 [==============================] - 0s 875us/step\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-08 11:04:06.707458: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    }
   ],
   "source": [
    "embed_out = embed_model.predict([unique_naics_x, unique_naics_x_naics])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "f622987d-4206-494f-8482-08ac99fff2f9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1311, 8)"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embed_out.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "2193800a-3e35-460c-8ee0-576d460f53bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_df = pd.DataFrame(embed_out, index=unique_naics_x.index) \n",
    "embed_df.columns = [f'emb_{i:03d}' for i in range(len(embed_df.columns))]\n",
    "embed_df.reset_index(inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "5ba4cb2e-4321-4dd4-89cf-5b9b48d90c71",
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_df = pd.DataFrame(embed_out, index=unique_naics_x.index)  \n",
    "embed_df.columns = [f'emb_{i:03d}' for i in range(len(embed_df.columns))]\n",
    "embed_df = embed_df.reset_index() \\\n",
    "    .merge(business_data[['LoanNr_ChkDgt', 'NAICS', 'NAICS_orig']], how='left', on='LoanNr_ChkDgt') \\\n",
    "    .drop(columns='LoanNr_ChkDgt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "17f1e84b-2aec-4b6b-bd27-04a79008b2de",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<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>emb_000</th>\n",
       "      <th>emb_001</th>\n",
       "      <th>emb_002</th>\n",
       "      <th>emb_003</th>\n",
       "      <th>emb_004</th>\n",
       "      <th>emb_005</th>\n",
       "      <th>emb_006</th>\n",
       "      <th>emb_007</th>\n",
       "      <th>NAICS</th>\n",
       "      <th>NAICS_orig</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.720058</td>\n",
       "      <td>-0.519170</td>\n",
       "      <td>0.409407</td>\n",
       "      <td>0.175112</td>\n",
       "      <td>-0.000629</td>\n",
       "      <td>0.898163</td>\n",
       "      <td>0.617700</td>\n",
       "      <td>0.434311</td>\n",
       "      <td>1095.0</td>\n",
       "      <td>722410</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1.427286</td>\n",
       "      <td>-1.779123</td>\n",
       "      <td>1.212681</td>\n",
       "      <td>0.405218</td>\n",
       "      <td>0.291517</td>\n",
       "      <td>1.934087</td>\n",
       "      <td>0.435628</td>\n",
       "      <td>1.120715</td>\n",
       "      <td>398.0</td>\n",
       "      <td>332721</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.536821</td>\n",
       "      <td>-0.637489</td>\n",
       "      <td>0.255909</td>\n",
       "      <td>0.151857</td>\n",
       "      <td>0.093758</td>\n",
       "      <td>0.563031</td>\n",
       "      <td>0.429567</td>\n",
       "      <td>0.376434</td>\n",
       "      <td>0.0</td>\n",
       "      <td>445299</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    emb_000   emb_001   emb_002   emb_003   emb_004   emb_005   emb_006  \\\n",
       "0  0.720058 -0.519170  0.409407  0.175112 -0.000629  0.898163  0.617700   \n",
       "1  1.427286 -1.779123  1.212681  0.405218  0.291517  1.934087  0.435628   \n",
       "2  0.536821 -0.637489  0.255909  0.151857  0.093758  0.563031  0.429567   \n",
       "\n",
       "    emb_007   NAICS NAICS_orig  \n",
       "0  0.434311  1095.0     722410  \n",
       "1  1.120715   398.0     332721  \n",
       "2  0.376434     0.0     445299  "
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embed_df.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "6a8d3c99-24d1-404c-a867-b67f6dd6a4a9",
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_df.to_parquet(Path(setup.temp_path).joinpath('25_DATA_embeddings.parquet'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "52e382b5-e9dc-47a1-9bf4-6d0b6b4a0c61",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tf_p39",
   "language": "python",
   "name": "tf_p39"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}