07_state_of_art_model_redo.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "07_state_of_art_model_redo.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "wWUKYkpSGCoI"
      },
      "source": [
        "!pip install -Uqq fastbook\n",
        "import fastbook\n",
        "fastbook.setup_book()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Rc-agrJSfwJP"
      },
      "source": [
        "\n",
        "#hide\n",
        "from fastbook import *"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cjg_77r2fxk4"
      },
      "source": [
        "from fastai.vision.all import *\n",
        "path = untar_data(URLs.IMAGENETTE)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wRHSJwYAnARp",
        "outputId": "ec271b6a-d8f9-411b-e10d-7f4a7924af47"
      },
      "source": [
        "im_stats = [*imagenet_stats]\n",
        "im_stats"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/plain": [
              "[[0.485, 0.456, 0.406], [0.229, 0.224, 0.225]]"
            ]
          },
          "execution_count": 4,
          "metadata": {},
          "output_type": "execute_result"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I8_em2wLmSye"
      },
      "source": [
        "def get_imagenette_dls(bs, size):\n",
        "  imnet = DataBlock(blocks=(ImageBlock, CategoryBlock),\n",
        "                    get_items = get_image_files,\n",
        "                    get_y = parent_label,\n",
        "                    item_tfms = Resize(460),\n",
        "                    batch_tfms = [*aug_transforms(size=size, min_scale=0.75),\n",
        "                                  Normalize.from_stats(*imagenet_stats)]\n",
        "  )\n",
        "\n",
        "  return imnet.dataloaders(path, bs=bs)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NAeWbwwonlKz",
        "outputId": "23e33ab5-c784-44dd-c36c-ad4794e527cf"
      },
      "source": [
        "path"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/plain": [
              "Path('/root/.fastai/data/imagenette2')"
            ]
          },
          "execution_count": 6,
          "metadata": {},
          "output_type": "execute_result"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "S7VuCU27oGj0",
        "outputId": "e3c5170b-181c-4de4-9ee7-1a30bf00c730"
      },
      "source": [
        "dls = get_imagenette_dls(64, 224)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/_tensor.py:1023: UserWarning: torch.solve is deprecated in favor of torch.linalg.solveand will be removed in a future PyTorch release.\n",
            "torch.linalg.solve has its arguments reversed and does not return the LU factorization.\n",
            "To get the LU factorization see torch.lu, which can be used with torch.lu_solve or torch.lu_unpack.\n",
            "X = torch.solve(B, A).solution\n",
            "should be replaced with\n",
            "X = torch.linalg.solve(A, B) (Triggered internally at  /pytorch/aten/src/ATen/native/BatchLinearAlgebra.cpp:760.)\n",
            "  ret = func(*args, **kwargs)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 354
        },
        "id": "LnjqwwJhnnqM",
        "outputId": "b63d7993-1081-4761-b18a-c1c6c9bc47ce"
      },
      "source": [
        "# train a blank model \n",
        "model = xresnet18(n_out=dls.c)\n",
        "learn = Learner(dls, model, loss_func=CrossEntropyLossFlat(), metrics=accuracy)\n",
        "learn.lr_find()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at  /pytorch/c10/core/TensorImpl.h:1156.)\n",
            "  return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "SuggestedLRs(valley=0.00015848931798245758)"
            ]
          },
          "execution_count": 8,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "id": "VS7xmTjIoFLX",
        "outputId": "3bb0dd1d-b1f3-4b05-ab35-2d158bf49500"
      },
      "source": [
        "# fit the entire model for 5 epochs with the new lr\n",
        "learn.fit_one_cycle(5, 3e-4)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>1.841481</td>\n",
              "      <td>1.634431</td>\n",
              "      <td>0.427931</td>\n",
              "      <td>03:20</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>1.360799</td>\n",
              "      <td>1.250425</td>\n",
              "      <td>0.603809</td>\n",
              "      <td>03:21</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>1.072050</td>\n",
              "      <td>0.931889</td>\n",
              "      <td>0.697535</td>\n",
              "      <td>03:21</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>3</td>\n",
              "      <td>0.919540</td>\n",
              "      <td>0.824963</td>\n",
              "      <td>0.748320</td>\n",
              "      <td>03:20</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>4</td>\n",
              "      <td>0.848013</td>\n",
              "      <td>0.783724</td>\n",
              "      <td>0.756908</td>\n",
              "      <td>03:21</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fQYTbjQZrR38"
      },
      "source": [
        "## Progressive Resizing"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 300
        },
        "id": "H2bidpNWpK55",
        "outputId": "c3db172a-c26f-4b70-f664-3f88a2045881"
      },
      "source": [
        "# implement progressive resizing with the same model and see if it is better\n",
        "dls = get_imagenette_dls(128, 128)\n",
        "model_2 = xresnet50(n_out=dls.c)\n",
        "learn = Learner(dls, model_2, loss_func=CrossEntropyLossFlat(), metrics=accuracy)\n",
        "learn.lr_find()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "SuggestedLRs(valley=0.0002290867705596611)"
            ]
          },
          "execution_count": 10,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 142
        },
        "id": "1c6ul7gCq6pm",
        "outputId": "5ab89350-5943-44dd-ab76-b6fe61f5bd48"
      },
      "source": [
        "learn.fit_one_cycle(3, 5e-3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>1.589526</td>\n",
              "      <td>1.823611</td>\n",
              "      <td>0.503734</td>\n",
              "      <td>03:03</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>1.223397</td>\n",
              "      <td>1.343979</td>\n",
              "      <td>0.578417</td>\n",
              "      <td>03:04</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>0.900740</td>\n",
              "      <td>0.760638</td>\n",
              "      <td>0.751307</td>\n",
              "      <td>03:05</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 300
        },
        "id": "7b0gXAjbq9mC",
        "outputId": "d9d5d4f9-7803-4ecd-946d-602b420391e5"
      },
      "source": [
        "# increase image size and decrease batch size\n",
        "dls = get_imagenette_dls(64, 224)\n",
        "learn.lr_find()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "SuggestedLRs(valley=0.0004786300996784121)"
            ]
          },
          "execution_count": 12,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 205
        },
        "id": "yuBxk4YzrLnn",
        "outputId": "3ae3f517-0f05-41f6-84d6-d9fa27a5f169"
      },
      "source": [
        "# apply transfer learning approach \n",
        "learn.fine_tune(3, 5e-4)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>0.766911</td>\n",
              "      <td>0.811830</td>\n",
              "      <td>0.751307</td>\n",
              "      <td>03:04</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>0.713242</td>\n",
              "      <td>0.694399</td>\n",
              "      <td>0.781553</td>\n",
              "      <td>03:04</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>0.655954</td>\n",
              "      <td>0.665772</td>\n",
              "      <td>0.785287</td>\n",
              "      <td>03:05</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>0.625331</td>\n",
              "      <td>0.637427</td>\n",
              "      <td>0.795743</td>\n",
              "      <td>03:08</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FesWc6ZfroYp"
      },
      "source": [
        "## TTA"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "id": "tez9PuLxrbXs",
        "outputId": "91585ee1-53b5-426f-def4-d2dd9bcf8fe7"
      },
      "source": [
        "# see if we get better accuracy with test time augmentation on the valdiation set\n",
        "preds, targs = learn.tta()\n",
        "accuracy(preds, targs)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "        <style>\n",
              "            /* Turns off some styling */\n",
              "            progress {\n",
              "                /* gets rid of default border in Firefox and Opera. */\n",
              "                border: none;\n",
              "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
              "                background-size: auto;\n",
              "            }\n",
              "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
              "                background: #F44336;\n",
              "            }\n",
              "        </style>\n",
              "      <progress value='0' class='' max='3' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      \n",
              "    </div>\n",
              "    \n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "TensorBase(0.8077)"
            ]
          },
          "execution_count": 14,
          "metadata": {},
          "output_type": "execute_result"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S560iQjFr7xl"
      },
      "source": [
        "## Mixup and Label Smoothing\n",
        "- train a model from scratch using mixup, another using label smoothing, and another combining both\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 173
        },
        "id": "ctk26kPyrxr2",
        "outputId": "c3b96e4e-ab2e-4463-db74-7b4281b7440f"
      },
      "source": [
        "learn = Learner(dls, xresnet50(n_out=dls.c), loss_func=CrossEntropyLossFlat(), metrics=accuracy, cbs=MixUp())\n",
        "learn.fit_one_cycle(4, 3e-3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>1.915911</td>\n",
              "      <td>2.345965</td>\n",
              "      <td>0.399552</td>\n",
              "      <td>05:55</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>1.633609</td>\n",
              "      <td>1.235714</td>\n",
              "      <td>0.581777</td>\n",
              "      <td>05:57</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>1.386902</td>\n",
              "      <td>0.884044</td>\n",
              "      <td>0.717326</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>3</td>\n",
              "      <td>1.216197</td>\n",
              "      <td>0.759218</td>\n",
              "      <td>0.772591</td>\n",
              "      <td>05:57</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "id": "B7hdk4kAwdwd",
        "outputId": "ab658e61-8538-4edf-a381-d22cdca9d369"
      },
      "source": [
        "# see if we get better accuracy with test time augmentation on the valdiation set after mixup\n",
        "preds, targs = learn.tta()\n",
        "accuracy(preds, targs)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "        <style>\n",
              "            /* Turns off some styling */\n",
              "            progress {\n",
              "                /* gets rid of default border in Firefox and Opera. */\n",
              "                border: none;\n",
              "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
              "                background-size: auto;\n",
              "            }\n",
              "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
              "                background: #F44336;\n",
              "            }\n",
              "        </style>\n",
              "      <progress value='0' class='' max='4' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      \n",
              "    </div>\n",
              "    \n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "TensorBase(0.7898)"
            ]
          },
          "execution_count": 16,
          "metadata": {},
          "output_type": "execute_result"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zSi_AAc-vswp"
      },
      "source": [
        "- To see better results with MixUp, we likely need to train for more epochs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 313
        },
        "id": "c1QLu5YQwRpN",
        "outputId": "a16abd24-9acb-4a53-81e4-f73fc887931b"
      },
      "source": [
        "learn = Learner(dls, xresnet50(n_out=dls.c), loss_func=LabelSmoothingCrossEntropy(), metrics=accuracy)\n",
        "learn.fit_one_cycle(4, 3e-3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "        <style>\n",
              "            /* Turns off some styling */\n",
              "            progress {\n",
              "                /* gets rid of default border in Firefox and Opera. */\n",
              "                border: none;\n",
              "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
              "                background-size: auto;\n",
              "            }\n",
              "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
              "                background: #F44336;\n",
              "            }\n",
              "        </style>\n",
              "      <progress value='2' class='' max='4' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      50.00% [2/4 11:53<11:53]\n",
              "    </div>\n",
              "    \n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>1.759637</td>\n",
              "      <td>1.773289</td>\n",
              "      <td>0.474234</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>1.401663</td>\n",
              "      <td>1.485612</td>\n",
              "      <td>0.617998</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table><p>\n",
              "\n",
              "    <div>\n",
              "        <style>\n",
              "            /* Turns off some styling */\n",
              "            progress {\n",
              "                /* gets rid of default border in Firefox and Opera. */\n",
              "                border: none;\n",
              "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
              "                background-size: auto;\n",
              "            }\n",
              "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
              "                background: #F44336;\n",
              "            }\n",
              "        </style>\n",
              "      <progress value='108' class='' max='167' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      64.67% [108/167 03:26<01:52 1.2608]\n",
              "    </div>\n",
              "    "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>1.759637</td>\n",
              "      <td>1.773289</td>\n",
              "      <td>0.474234</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>1.401663</td>\n",
              "      <td>1.485612</td>\n",
              "      <td>0.617998</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>1.180420</td>\n",
              "      <td>1.209459</td>\n",
              "      <td>0.726288</td>\n",
              "      <td>05:55</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>3</td>\n",
              "      <td>1.046575</td>\n",
              "      <td>1.015919</td>\n",
              "      <td>0.813294</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "id": "Dga9ehlLwe2Z",
        "outputId": "c3822deb-eaba-4080-a325-8234095d2bf5"
      },
      "source": [
        "# see if we get better accuracy with test time augmentation on the valdiation set after label smoothing\n",
        "preds, targs = learn.tta()\n",
        "accuracy(preds, targs)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "        <style>\n",
              "            /* Turns off some styling */\n",
              "            progress {\n",
              "                /* gets rid of default border in Firefox and Opera. */\n",
              "                border: none;\n",
              "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
              "                background-size: auto;\n",
              "            }\n",
              "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
              "                background: #F44336;\n",
              "            }\n",
              "        </style>\n",
              "      <progress value='0' class='' max='4' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      \n",
              "    </div>\n",
              "    \n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "TensorBase(0.8163)"
            ]
          },
          "execution_count": 18,
          "metadata": {},
          "output_type": "execute_result"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UvlwZAOswf3K"
      },
      "source": [
        "- Combine both to see how it affects model results"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "background_save": true
        },
        "id": "f-dAfCXjwuvI",
        "outputId": "c2b7d201-9f70-40dc-db90-f3c0b7fd8f93"
      },
      "source": [
        "learn = Learner(dls, xresnet50(n_out=dls.c), loss_func=LabelSmoothingCrossEntropy(), metrics=accuracy, cbs=MixUp())\n",
        "learn.fit_one_cycle(4, 3e-3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>2.010575</td>\n",
              "      <td>4.006091</td>\n",
              "      <td>0.312547</td>\n",
              "      <td>05:57</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>1.774568</td>\n",
              "      <td>1.920753</td>\n",
              "      <td>0.463779</td>\n",
              "      <td>05:56</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>1.577706</td>\n",
              "      <td>1.261139</td>\n",
              "      <td>0.703883</td>\n",
              "      <td>05:57</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>3</td>\n",
              "      <td>1.463302</td>\n",
              "      <td>1.110149</td>\n",
              "      <td>0.764003</td>\n",
              "      <td>05:55</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "background_save": true
        },
        "id": "DIcywVXRwzqq",
        "outputId": "809a74c1-7f2f-4a6b-bb4f-2be454cf57ad"
      },
      "source": [
        "# see if we get better accuracy with test time augmentation on the valdiation set after ixup and label smoothing\n",
        "preds, targs = learn.tta()\n",
        "accuracy(preds, targs)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "        <style>\n",
              "            /* Turns off some styling */\n",
              "            progress {\n",
              "                /* gets rid of default border in Firefox and Opera. */\n",
              "                border: none;\n",
              "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
              "                background-size: auto;\n",
              "            }\n",
              "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
              "                background: #F44336;\n",
              "            }\n",
              "        </style>\n",
              "      <progress value='0' class='' max='4' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      \n",
              "    </div>\n",
              "    \n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "TensorBase(0.7812)"
            ]
          },
          "execution_count": null,
          "metadata": {},
          "output_type": "execute_result"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-hcqSznVx-TF"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}