CXR_SIMCLR_Pretrainied.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Joyatulya/colab_notebooks/blob/main/CXR_SIMCLR_Pretrainied.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "tasplczpf_Xf"
      },
      "outputs": [],
      "source": [
        "!nvidia-smi"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "3IhUpi6Bn-9x"
      },
      "outputs": [],
      "source": [
        "# !sudo apt install -qq cadaver\n",
        "# username BIMCV-PadChest-FULL\n",
        "# !cadaver https://b2drop.bsc.es/public.php/webdav/"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oCb_4p0ce6SV"
      },
      "source": [
        "#Plan\n",
        "\n",
        "## How to approach this??\n",
        "\n",
        "I want to do a classification problem where my model is able to understand and make diagnosis on the basis of the data that it sees in the cxr\n",
        "\n",
        "Possible problems or hurdles?\n",
        "- I had that problem of generalisation in the previous attempt at this, the model did really well on the cxr that is saw and did poorly on everything else.\n",
        "\n",
        "  How can this be mitigated?? \n",
        "  - for starters I could make a labeled dataset from the ambala put dont really know that it will be worht the effort.\n",
        "  - I could get another random dataset from somewhere else which is already labeled and use that as the test dataset. What I think is that the model learns the different really small nuances inside of the pics which are common to the same localty and i need to be able to mitigate that.\n",
        "\n",
        "- Another thing is that I want to learn more about the CLIP paradigm of models as from the outside they seem really promising and my mind is being drawn there but i know that I do that quite a lot and that mihgt not be the best thing right now.\n",
        "\n",
        "## The more important problem\n",
        "\n",
        "I dont have setup how i will put in the labels from all these datasets and what is the best way to do so and I dont fucking know how to approach that.\n",
        "\n",
        "- Should I stick to my previous approach and agglomerate these different but similar labels together ot should I go for the new CLIP method.\n",
        "- Even if go for any of the above methods, my datasets are varied and how will I join them, ie one has reports and the other doesnt have reporsts and the one that has reports have them in spanish.\n",
        "\n",
        "## Some ideas\n",
        "1. I can use the simclrv2 to its completion and use that for the classification problema and then distill their knowledge into the smaller models where the classification problem will have a binary classification task and we will construct different models for all the different diseases like other people are doing.\n",
        "\n",
        "  One more thing that came ot my mind is that I can make  a model with freature representation and then use different classifier heads for different diseases  just like an adaptoer on top of the main backbone which will be really helpful.\n",
        "\n",
        "  There is one pitfall in this idea as it assumes that the backbone has learned accurate feature representations and that is not quite necessary, Moreover I want to think that should I only train the classifier head or should I tune the base model wiht it also. SHould read more about it in the papers.\n",
        "\n",
        "\n",
        "For now, lets read the paper that sat us on this path, the cxr semi-supervised paper and try to read a few more things from it and then get on with it. CLIP might be the key."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OYvP6YgRKgFG"
      },
      "source": [
        "#Pretraining Prep"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7-byMfPGgEFe"
      },
      "source": [
        "##Setting up Enviornment"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SOEPxs3reJT6",
        "outputId": "0e519b89-2177-4e71-b882-0d0280ec1a24"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "fatal: destination path 'utils' already exists and is not an empty directory.\n"
          ]
        }
      ],
      "source": [
        "!git clone -q  https://github.com/Joyatulya/utils.git\n",
        "!pip -qq install transformers pytorch-lightning wandb torchtyping\n",
        "# !pip -qq install -U rich\n",
        "# !pip install vissl"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Lgi3UjwKeWep",
        "outputId": "c7dbd75e-b8cc-4037-92e9-4f9351f959ef"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Copying Kaggle/wandb/git\n"
          ]
        }
      ],
      "source": [
        "from utils import enviornmet_setup\n",
        "enviornmet_setup.config_files()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Znyy-egEfEOh"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import pandas as pd\n",
        "import os\n",
        "import torch\n",
        "from pytorch_lightning.callbacks import ModelCheckpoint, RichProgressBar, LearningRateMonitor\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torchvision\n",
        "from torchvision import models\n",
        "from torch import optim\n",
        "import torchvision.transforms as T\n",
        "import torch.utils.data as data\n",
        "from torch.utils.data import Dataset, DataLoader\n",
        "import pytorch_lightning as pl\n",
        "import wandb  \n",
        "from glob import glob\n",
        "from torchvision import io\n",
        "import h5py\n",
        "import shutil\n",
        "import pathlib\n",
        "from pytorch_lightning.loggers import WandbLogger\n",
        "import cv2\n",
        "from tqdm.auto import tqdm\n",
        "from torchvision.io.image import ImageReadMode\n",
        "\n",
        "from sklearn.model_selection import train_test_split"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "mKcwepGg2wqu",
        "outputId": "da023fcc-bfc0-4429-8817-0d6aa1e5fb98"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mjoy_atulya\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n",
            "Global seed set to 1996\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1996"
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ],
      "source": [
        "wandb.login()\n",
        "pl.seed_everything(1996)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "wakAfuwNmRXd"
      },
      "outputs": [],
      "source": [
        "CLINICAL_LABELS = ['PatientID', \"PatientBirth\",'PatientSex_DICOM', 'ViewPosition_DICOM','Pediatric','ReportID','Report','MethodLabel','Labels','Localizations','LabelsLocalizationsBySentence','labelCUIS', 'LocalizationsCUIS']\n",
        "HOME_DIR = pathlib.Path().cwd() / 'drive' / 'MyDrive' / 'adv_cxr'"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Jpc7NFVrgQp3"
      },
      "source": [
        "##Data Preparation"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tnJ1DU4ihp7l",
        "outputId": "48c0913d-ea53-42c3-ce3a-8769ffe18260"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/IPython/core/interactiveshell.py:2882: DtypeWarning: Columns (5,20,21) have mixed types.Specify dtype option on import or set low_memory=False.\n",
            "  exec(code_obj, self.user_global_ns, self.user_ns)\n"
          ]
        }
      ],
      "source": [
        "pad_df = pd.read_csv( HOME_DIR / 'padchest' / 'padchest.csv')\n",
        "nih_df = pd.read_csv( HOME_DIR / 'nih' / 'nih.csv')\n",
        "vinbig_df = pd.read_csv( HOME_DIR / 'vinbig' / 'vinbig.csv')\n",
        "tb_df = pd.read_csv( HOME_DIR / 'tb' / 'tb.csv')\n",
        "combined_df = pd.read_csv(HOME_DIR / 'combined_dataset.csv')\n",
        "assert len(pad_df) + len(nih_df) + len(vinbig_df) + len(tb_df) == len(combined_df), 'Length of all dataframes dont add up'\n",
        "# combined_df.rename({'max' : 'max_value'}, axis = 1, inplace = True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P6tqstiJo983"
      },
      "source": [
        "### BIMCV df"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "EGyYywxEeeT7"
      },
      "outputs": [],
      "source": [
        "labels = pad_df.Labels.values\n",
        "arrayer = np.vectorize(lambda x : ast.literal_eval(x), otypes = ['str'])\n",
        "# arrayer(labels)\n",
        "labels\n",
        "# pad_df.Labels.isna()."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "yGp-ngin5Ejh"
      },
      "outputs": [],
      "source": [
        "# items_in_h5 = []\n",
        "# # for key, val in padchest_ds.imgs.items():\n",
        "#   # items_in_h5.append((key, len(val)))\n",
        "# for i in range(51):\n",
        "#   print(padchest_ds.img_names[str(i)], padchest_ds.imgs[str(i)])\n",
        "# padchest_ds.img_names, padchest_ds.imgs"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "KXUJPqxQfO5w"
      },
      "outputs": [],
      "source": [
        "# for file_num in range(51):\n",
        "#   if not pad_df[pad_df.file == file_num].shape == pad_df[pad_df.ImageDir == file_num].shape:\n",
        "#     print(pad_df[pad_df.ImageDir == file_num].shape)\n",
        "#     print(pad_df[pad_df.file == file_num].shape)\n",
        "#     print(file_num)\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "PGxPLSv6d2ei"
      },
      "outputs": [],
      "source": [
        "# items_in_text_file = []\n",
        "# for num , f in enumerate((HOME_DIR / 'file_check').glob('*.txt')):\n",
        "#   with open(f,'r') as file:\n",
        "#     x = file.readlines()\n",
        "#     items_in_text_file.append((num, x[-1]))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "UzObEuFT8D1Z"
      },
      "outputs": [],
      "source": [
        "# np.array((items_in_text_file, items_in_h5, pad_df.ImageDir.value_counts(sort = False)))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "EQ-Zb9xnmMl2"
      },
      "outputs": [],
      "source": [
        "# from collections import Counter\n",
        "# all_unique_labels = raw_df.Labels.values\n",
        "# # '|'.join(all_unique_labels.flatten())\n",
        "# all_labels = []\n",
        "# for label in all_unique_labels.flatten():\n",
        "#   try:\n",
        "#     label = label.replace('[','')\n",
        "#     label = label.replace(']','')\n",
        "#     label = label.replace(\"'\",\"\")\n",
        "#     label = label.split(',')\n",
        "#     for word in label:\n",
        "#       word = word.lstrip()\n",
        "#       all_labels.append(word)\n",
        "#     # break\n",
        "#   except:\n",
        "#     all_labels.append('ERROR')\n",
        "# unique_vals = Counter(all_labels)\n",
        "# unique_vals.most_common()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "3btbJwBBRPMO"
      },
      "outputs": [],
      "source": [
        "def write_to_file(data):\n",
        "  with open('log_padchest.txt', 'a') as f:\n",
        "    f.write(data + '\\n')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "kemrNq6a8Cky"
      },
      "outputs": [],
      "source": [
        "from typing import Optional\n",
        "from torchtyping import TensorType\n",
        "class BaseDataset(data.Dataset):\n",
        "  \"\"\"\n",
        "  Maybe can also add a small buffer instead of a single file which can be used.\n",
        "  \"\"\"\n",
        "  def __init__(self,df, base_dir):\n",
        "    \"\"\"\n",
        "    df : Pandas df which have all the data for our dataset\n",
        "    base_dir: dir in which we have .h5 files of the data\n",
        "    \"\"\"\n",
        "    super().__init__()\n",
        "    self.df = df\n",
        "    self.file = df.file.values\n",
        "    self.file_index = df.file_index.values\n",
        "    self.base_dir = base_dir\n",
        "    self.h5_files = list(base_dir.glob('*.h5'))\n",
        "    self.imgs, self.img_names = self._load_h5_files()\n",
        "    assert len(self.imgs) == len(self.img_names), f\"Unequal arrays inside of h5 file\"\n",
        "\n",
        "  def __len__(self):\n",
        "    # You should check here if this is the reason of the index error\n",
        "    return len(self.df)\n",
        "\n",
        "  def _load_h5_files(self):\n",
        "    \"\"\"\n",
        "    Loads the generator of images and returns a dict\n",
        "    \"\"\" \n",
        "    imgs = {}\n",
        "    names = {}\n",
        "    for file in self.h5_files:\n",
        "      data = h5py.File(file,)\n",
        "      _imgs = data['img']\n",
        "      _names = data['name']\n",
        "      name = file.name.split('.')[0]\n",
        "      imgs.update({name : _imgs})\n",
        "      names.update({name : _names})\n",
        "    return imgs, names\n",
        "    \n",
        "## PADCHEST Dataset ##\n",
        "class PadchestDataset(BaseDataset):\n",
        "  def __init__(self, df, base_dir):\n",
        "    super().__init__(df, base_dir)\n",
        "    self.imageid = df.ImageID.values\n",
        "    \n",
        "  def __getitem__(self, idx):\n",
        "    file_name = str(self.file[idx])\n",
        "    name_h5 = self.img_names[file_name][self.file_index[idx]].decode()\n",
        "    name_df = self.imageid[idx]\n",
        "\n",
        "    try:\n",
        "      img = self.imgs[file_name][self.file_index[idx]]\n",
        "    except IndexError as e:\n",
        "      print(file_name)\n",
        "      print(name_h5, name_df)\n",
        "      write_to_file(f\"{file_name} = {name_h5} - {name_df}\")\n",
        "      # raise e\n",
        "      pass\n",
        "\n",
        "    try:\n",
        "      assert name_h5 == name_df, f\"{name_h5} not equal to {name_df}, in file number {file_name}\"\n",
        "    except AssertionError as e:\n",
        "      write_to_file(f\"{file_name} = {name_h5} - {name_df}\")\n",
        "      pass\n",
        "\n",
        "    return torch.tensor(img, dtype = torch.float)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "DOP5SRu04__Y"
      },
      "outputs": [],
      "source": [
        ")\n",
        "total_files = 0\n",
        "all_files_df = pd.DataFrame(columns=['image_name', 'file_name', 'index_in_file', 'max'])\n",
        "# all_files = list(HOME_DIR.glob('*/*.h5'))\n",
        "decoder = np.vectorize(lambda x : x.decode())\n",
        "for file_num, file_name in tqdm(enumerate(all_files), total = len(all_files)):\n",
        "  with h5py.File(file_name,'r') as file:\n",
        "    imgs = file['img']\n",
        "    names = file['name']\n",
        "    names =  decoder(names)\n",
        "    df = pd.DataFrame({\n",
        "        'image_name' : np.array(names)\n",
        "    })\n",
        "    imgs = np.array(imgs)\n",
        "    df['max'] = np.max(np.max(imgs, 2),-1).flatten()\n",
        "  df['file_name'] = file_name\n",
        "  df['index_in_file'] = df.index\n",
        "  all_files_df = pd.concat((all_files_df, df), ignore_index = True)\n",
        "  # break\n",
        "  # total_files += names.shape[0]\n",
        "# df['file_nums'] = df.index\n",
        "all_files_df\n",
        "  # print(imgs[0].max())\n",
        "#   # print(file, imgs.shape, names.shape)\n",
        "\n",
        "# fif = h5py.File(all_files[])\n",
        "# imgs = fif['img']\n",
        "# imgs[-1][0].max()\n",
        "# plt.imshow(h5py.File(all_files[50])['img'][0][0])\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "SqUYqjLsiO09"
      },
      "outputs": [],
      "source": [
        "# all_files_df['h5_name'] = all_files_df.file_name\n",
        "# all_files_df['h5_name'] = all_files_df['h5_name'].apply(lambda x : x.name)\n",
        "# all_files_df.to_csv(HOME_DIR / 'combined_dataset.csv')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "t6_d-l-dUJqC"
      },
      "outputs": [],
      "source": [
        "all_files_df = pd.read_csv(HOME_DIR / 'combined_dataset.csv')\n",
        "class CombinedDataset(Dataset):\n",
        "\n",
        "  def __init__(self, df):\n",
        "    super().__init__()\n",
        "    self.df = df\n",
        "    self.h5_names = df.h5_name.values\n",
        "    self.index_in_file = df.index_in_file.values\n",
        "    self.max_sizes = df['max'].values\n",
        "    self.h5_files = HOME_DIR.glob('*/*.h5')\n",
        "    self.imgs, self.img_names = self._load_h5_files()\n",
        "    self.transform = T.Compose([\n",
        "          # T.RandomPosterize(3, .2),\n",
        "          # T.RandomRotation(40, fill = 0),\n",
        "          T.GaussianBlur(7),\n",
        "          # T.RandomAutocontrast(0.5),\n",
        "          T.RandomResizedCrop(284, (.35,1)),\n",
        "          # T.RandomPerspective(.1),\n",
        "          # T.RandomEqualize(0.2),\n",
        "          T.ConvertImageDtype(torch.float),\n",
        "        T.Normalize(116.549, 72.247)\n",
        "      ])\n",
        "  def __len__(self):\n",
        "    return len(self.df)\n",
        "\n",
        "  def __getitem__(self, idx):\n",
        "    img = self.get_image(idx)\n",
        "    if self.max_sizes[idx] > 300:\n",
        "      img = (img * 255) / 65536\n",
        "    img = torch.tensor(img, dtype = torch.uint8)\n",
        "    img_1 = self.transform(img)\n",
        "    img_2 = self.transform(img)\n",
        "    return img_1, img_2\n",
        "\n",
        "  def _load_h5_files(self):\n",
        "    \"\"\"\n",
        "    Loads the generator of images and returns a dict\n",
        "    \"\"\" \n",
        "    imgs = {}\n",
        "    names = {}\n",
        "    for file in self.h5_files:\n",
        "      data = h5py.File(file,)\n",
        "      _imgs = data['img']\n",
        "      _names = data['name']\n",
        "      name = file.name\n",
        "      imgs.update({name : _imgs})\n",
        "      names.update({name : _names})\n",
        "    return imgs, names\n",
        "\n",
        "  def get_image(self, idx):\n",
        "    file_name = self.h5_names[idx]\n",
        "    index_in_file = self.index_in_file[idx]\n",
        "    return self.imgs[file_name][index_in_file]\n",
        "combined_ds = CombinedDataset(all_files_df)\n",
        "combined_dl = DataLoader(combined_ds, 1024, num_workers = 8)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "RsBr-FzYGQhd"
      },
      "outputs": [],
      "source": [
        "TOTAL_FILES = 288781"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "kqsUrKwF8eSk"
      },
      "outputs": [],
      "source": [
        "## NIHDataset ##seuj;k\n",
        "class NIHDataset(BaseDataset):\n",
        "  def __init__(self, df, base_dir):\n",
        "    super().__init__(df, base_dir)\n",
        "    self.imageid = df.image_id.values\n",
        "    self.transform = T.Compose([\n",
        "        T.ToTensor(),\n",
        "        # T.Normalize(119.104, 68.831)\n",
        "    ])\n",
        "    \n",
        "  def __getitem__(self, idx):\n",
        "    file_name = str(self.file[idx])\n",
        "    img = self.imgs[file_name][self.file_index[idx]]\n",
        "    img = self.transform(img)\n",
        "    img= img.moveaxis(1,0).transpose(1,2)\n",
        "    img = torch.cat((img,img,img), 0)\n",
        "    return img\n",
        "\n",
        "vinbig_ds = NIHDataset(vinbig_df, HOME_DIR / 'vinbig')\n",
        "vinbig_dl = DataLoader(vinbig_ds, 1024)\n",
        "imgs = next(iter(vinbig_dl))\n",
        "plt.imshow(imgs[0,0,:])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "EFv72C02_NoO"
      },
      "outputs": [],
      "source": [
        "class AmbalaDataset(Dataset):\n",
        "  def __init__(self, base_dir):\n",
        "    self.base_dir = base_dir\n",
        "    self.img_list = list(base_dir.glob('*.jpeg'))\n",
        "    self.transform = T.Compose([\n",
        "        T.Resize((284,284)),\n",
        "        T.Normalize(199.104, 69.8)\n",
        "    ])\n",
        "  def __len__(self):\n",
        "    return len(self.img_list)\n",
        "  \n",
        "  def __getitem__(self, idx):\n",
        "    img = io.read_image(str(self.img_list[idx]), mode = ImageReadMode.GRAY).float()\n",
        "    img = self.transform(img)\n",
        "    return img\n",
        "\n",
        "ambala_ds = AmbalaDataset(pathlib.Path('./drive/MyDrive/tb_ambala_cxr'))\n",
        "ambala_dl = DataLoader(ambala_ds, 32)\n",
        "next(iter(ambala_dl)).shape"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "LiesCfXpJma6"
      },
      "outputs": [],
      "source": [
        "mean = []\n",
        "std = []\n",
        "for img in tqdm(combined_dl, total = len(combined_dl)):\n",
        "  mean.append(img.mean())\n",
        "  std.append(img.std())\n",
        "  # break"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "JGK9y67LYOh-"
      },
      "outputs": [],
      "source": [
        "np.mean(mean), np.mean(std)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mRMDIrcWngrw"
      },
      "source": [
        "##NIH"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "GNMfpMJ4nmrw"
      },
      "outputs": [],
      "source": [
        "PATHOLOGY_LIST = ['Cardiomegaly','Emphysema','Effusion',\n",
        "                  'Pneumothorax','Mass','Edema','Consolidation',\n",
        "                  'Fibrosis',]\n",
        "                  \n",
        "def standardise_df(df):\n",
        "  working_df = df.copy()\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Infiltration','Consolidation')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Pneumonia','Consolidation')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Atelectasis','Consolidation')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Nodule/Mass','Mass')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Nodule','Mass')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Lung Opacity','Mass')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Pulmonary fibrosis','Fibrosis')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('ILD','Fibrosis')\n",
        "  working_df['findings'] = working_df['findings'].str.replace('Pleural effusion', 'Effusion')\n",
        "  \n",
        "  return working_df"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "W8mYkf91oHXB"
      },
      "outputs": [],
      "source": [
        "std_nih_df = standardise_df(nih_df)\n",
        "std_nih_df['Effusion'] = std_nih_df.findings.str.contains('Effusion').astype('float')\n",
        "std_nih_df = pd.merge(std_nih_df, combined_df, on = 'image_id')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GV7sm_jRgble"
      },
      "source": [
        "#Training"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "D-qeSEACpNiT"
      },
      "outputs": [],
      "source": [
        "from torchvision.transforms.transforms import ToTensor\n",
        "class ClassificationDataset(Dataset):\n",
        "\n",
        "  def __init__(self, df, return_id = False):\n",
        "    '''\n",
        "    return_id returns the id of images in the geT item method\n",
        "    '''\n",
        "    super().__init__()\n",
        "    self.df = df\n",
        "    self.image_ids = df.image_id.values\n",
        "    # For retrieving files\n",
        "    self.h5_names = df.h5_name.values\n",
        "    self.index_in_file = df.index_in_file.values\n",
        "    self.max_sizes = df['max'].values\n",
        "    self.h5_files = HOME_DIR.glob('*/*.h5')\n",
        "    self.imgs, self.img_names = self._load_h5_files()\n",
        "    self.transform = T.Compose([\n",
        "      T.ToTensor(),\n",
        "        T.Normalize(116.549, 72.247)\n",
        "      ])\n",
        "    self.labels = self.df.Effusion.values\n",
        "    self.return_id = return_id\n",
        "    \n",
        "  def __len__(self):\n",
        "    return len(self.df)\n",
        "\n",
        "  def __getitem__(self, idx):\n",
        "    img = self.get_image(idx)\n",
        "    img = self.transform(img)\n",
        "    img = img.moveaxis(1,0)\n",
        "    img = img.transpose(1,2)\n",
        "    label = self.labels[idx]\n",
        "    if self.return_id: return self.image_ids[idx], img, label\n",
        "    return img, label\n",
        "\n",
        "  def _load_h5_files(self):\n",
        "    \"\"\"\n",
        "    Loads the generator of images and returns a dict\n",
        "    \"\"\" \n",
        "    imgs = {}\n",
        "    names = {}\n",
        "    for file in self.h5_files:\n",
        "      data = h5py.File(file)\n",
        "      _imgs = data['img']\n",
        "      _names = data['name']\n",
        "      name = file.name\n",
        "      imgs.update({name : _imgs})\n",
        "      names.update({name : _names})\n",
        "    return imgs, names\n",
        "\n",
        "  def get_image(self, idx):\n",
        "    file_name = self.h5_names[idx]\n",
        "    index_in_file = self.index_in_file[idx]\n",
        "    return self.imgs[file_name][index_in_file]\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "NmXo5iSTvDXK"
      },
      "outputs": [],
      "source": [
        "from utils.architecture import simclr\n",
        "simclr = simclr.SimCLR(.07, 1.32, 1e-4, 4000)\n",
        "simclr = simclr.load_from_checkpoint(HOME_DIR / 'models'/ 'simclr' / 'full' / 'regnet_8_full' / '2jmq032w' / 'checkpoints' / 'simclr-regnet_8_full-epoch=10-val_loss=0.02-val_acc_top5=1.00.ckpt')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9Nly4jhAwROe",
        "outputId": "fb91448d-d8d3-43c5-f238-898da8424a67"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Linear(in_features=1920, out_features=1024, bias=True)"
            ]
          },
          "metadata": {},
          "execution_count": 11
        }
      ],
      "source": [
        "backbone_model = simclr.encoder\n",
        "backbone_model.requires_grad_(False)\n",
        "# backbone_model.stem = simclr.encoder.stem\n",
        "# simclr\n",
        "backbone_model.fc.requires_grad_(True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "XbJyDJE-sLGo"
      },
      "outputs": [],
      "source": [
        "import torchmetrics as metrics\n",
        "# from sklearn import metrics\n",
        "import copy\n",
        "class Classifier(pl.LightningModule):\n",
        "  \n",
        "  def __init__(self,lr, hidden_dim, num_classes,pos_weights, dropout = .1):\n",
        "    super().__init__()\n",
        "    self.save_hyperparameters()\n",
        "    print(\"\\nPositive Weights for the loss are ->\", pos_weights, end = '\\n')\n",
        "    self.backbone =copy.deepcopy(backbone_model)\n",
        "    self.classifier = nn.Sequential(\n",
        "        nn.ReLU(True),\n",
        "        nn.Linear(1024, hidden_dim),\n",
        "        nn.ReLU(True),\n",
        "        nn.Linear(hidden_dim, num_classes)\n",
        "    )\n",
        "    self.avg_prc = metrics.AveragePrecision()\n",
        "    self.recall = metrics.Recall(num_classes = num_classes)\n",
        "    self.prec = metrics.Precision(num_classes=num_classes, )\n",
        "\n",
        "  def forward(self, x):\n",
        "    return self.classifier(self.backbone(x))\n",
        "\n",
        "  def training_step(self, batch, batch_idx):\n",
        "    return self._calculate_loss(batch, mode = 'train')\n",
        "    \n",
        "  def validation_step(self, batch, batch_idx):\n",
        "    self._calculate_loss(batch, mode = 'val')\n",
        "\n",
        "  def predict_step(self, batch, batch_idx):\n",
        "    # img_ids, image, labels = batch\n",
        "    # predictions = self(image)\n",
        "    # pred_df = pd.DataFrame(\n",
        "    #     {\n",
        "    #         'image_id' : img_ids,\n",
        "    #         'predictions' : predictions.flatten().detach()\n",
        "    #     }\n",
        "    # )\n",
        "    # return pred_df\n",
        "    imgs, labels = batch\n",
        "    predictions = self(imgs)  \n",
        "    return predictions.cpu().numpy().flatten()\n",
        "\n",
        "  def _calculate_loss(self, batch, mode = 'train'):\n",
        "    imgs, labels = batch\n",
        "    preds = self(imgs).view(-1)\n",
        "    loss = self.loss_fn(preds, labels)\n",
        "    prc = self.avg_prc(preds, labels.long())\n",
        "    precision = self.prec(preds, labels.long())\n",
        "    recall = self.recall(preds, labels.long())\n",
        "    self.log(f'{mode}_prc', prc, prog_bar = True)\n",
        "    self.log(f'{mode}_precision', precision, prog_bar = True)\n",
        "    self.log(f'{mode}_recall', recall, prog_bar = True)\n",
        "    self.log(f'{mode}_loss', loss, prog_bar = True)\n",
        "    \n",
        "    return loss\n",
        "\n",
        "  def loss_fn(self, preds, labels):\n",
        "    return nn.BCEWithLogitsLoss(pos_weight = self.hparams.pos_weights)(preds, labels)\n",
        "\n",
        "  def configure_optimizers(self):\n",
        "    optimizer = optim.Adam(self.parameters(), lr = self.hparams.lr)\n",
        "    return optimizer"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lLFEgUW0ckkA",
        "outputId": "988840e8-d860-4cc9-a832-fe3e8df40e35"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Positive Weights for the loss are -> tensor([1.], dtype=torch.float64)\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torchmetrics/utilities/prints.py:36: UserWarning: Metric `AveragePrecision` will save all targets and predictions in buffer. For large datasets this may lead to large memory footprint.\n",
            "  warnings.warn(*args, **kwargs)\n"
          ]
        }
      ],
      "source": [
        "model = Classifier.load_from_checkpoint('./drive/MyDrive/adv_cxr/models/classification/Effusion_2048_hidden_1024_samples_no_dropout/3cgj2gki/checkpoints/Effusion_2048_hidden_1024_samples_no_dropout-epoch=3-val_loss=0.63-val_acc_top5=0.00.ckpt')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hjmZvpMrTYQb"
      },
      "outputs": [],
      "source": [
        "num_samples = 13317 * 2\n",
        "path = std_nih_df[std_nih_df.Effusion == 1.].sample(n = num_samples // 2)\n",
        "non_path = std_nih_df[std_nih_df.Effusion == 0.].sample(n = num_samples // 2)\n",
        "final_df = pd.concat((path, non_path))\n",
        "final_df = final_df.sample(frac = 1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "FiGk3-9Mzedd"
      },
      "outputs": [],
      "source": [
        "from pytorch_lightning import callbacks as cb\n",
        "def train_simclr(model_name,batch_size, max_epochs=500, **kwargs):\n",
        "\n",
        "    MODEL_NAME = model_name\n",
        "    CHECKPOINT_PATH = HOME_DIR / 'models' / 'classification'\n",
        "    NUM_WORKERS = 8\n",
        "\n",
        "    # logger = pl.loggers.WandbLogger(\n",
        "    #     project = 'CXR_classification',\n",
        "    #   name = MODEL_NAME,\n",
        "    #   group = 'Effusion'    \n",
        "    # ) \n",
        "\n",
        "    model_checkpoint = ModelCheckpoint(\n",
        "                                 save_weights_only=True,\n",
        "                                 save_top_k = 3,\n",
        "                                 filename = f'{MODEL_NAME}' + '-{epoch}-{val_loss:.2f}-{val_prc:.2f}',  \n",
        "                                  mode='min',\n",
        "                                   monitor='val_loss'\n",
        "                                  )\n",
        "    \n",
        "    trainer = pl.Trainer(default_root_dir=CHECKPOINT_PATH,\n",
        "                          accelerator = 'auto',\n",
        "                          max_epochs=max_epochs,\n",
        "                          # logger = logger,\n",
        "                         callbacks=[\n",
        "                            # model_checkpoint,\n",
        "                             LearningRateMonitor('epoch'),\n",
        "                             cb.StochasticWeightAveraging(swa_lrs=1e-2)\n",
        "                             ],\n",
        "                        # fast_dev_run = True,\n",
        "                        #  num_sanity_val_steps=3,\n",
        "                        #  overfit_batches=10,\n",
        "                        #  log_every_n_steps = 5,\n",
        "                        #  val_check_interval = .5,\n",
        "                         precision = 16,\n",
        "                         \n",
        "                        #  auto_scale_batch_size=True\n",
        "                         )\n",
        "\n",
        "    trainer.logger._default_hp_metric = None # Optional logging argument that we don't need\n",
        "\n",
        "    vals = final_df.Effusion.value_counts().values\n",
        "    pos_weights = vals[0] / vals[1]\n",
        "    pos_weights = torch.tensor([pos_weights], device = torch.device('cuda:0'))\n",
        "    \n",
        "    train_df, val_df = train_test_split(final_df, test_size = .15)\n",
        "    train_ds = ClassificationDataset(train_df)\n",
        "    val_ds = ClassificationDataset(val_df)\n",
        "\n",
        "    train_dl = DataLoader(train_ds, batch_size = batch_size, shuffle = True, drop_last = True, pin_memory = True, num_workers = NUM_WORKERS)\n",
        "    val_dl = DataLoader(val_ds, batch_size = batch_size, shuffle = False, drop_last = False, pin_memory = True, num_workers = NUM_WORKERS)\n",
        "\n",
        "\n",
        "    model = Classifier(num_classes = 1, pos_weights = pos_weights,**kwargs)\n",
        "\n",
        "    try:\n",
        "      trainer.fit(\n",
        "        model,\n",
        "        train_dl,\n",
        "        val_dl\n",
        "       \n",
        "        )\n",
        "    except KeyboardInterrupt:\n",
        "      pass\n",
        "    wandb.finish()\n",
        "    model = model.load_from_checkpoint(trainer.checkpoint_callback.best_model_path) # Load best checkpoint after training\n",
        "\n",
        "    preds = trainer.predict(model, val_dl) \n",
        "    preds = np.array(preds)\n",
        "    preds = np.hstack(preds)\n",
        "    val_df['predictions']  = preds\n",
        "    # pred_df = pd.merge(val)\n",
        "    return model, val_df"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 664,
          "referenced_widgets": [
            "35272b94d6bd46509e6fe9cc82280469",
            "8205ae282d8747f1bfa12729ab908e27",
            "f2cde6a705624a87a344e1adb92be88d",
            "13252b4efb704551a383696b7251beed",
            "aa10847a99f74929b601c6252747d86f",
            "2709658a7ce9477fa78c5ffbfa922178",
            "998217366fdc40adaa7aa014ce70754c",
            "83bda3c98e2f491b935ce42904a4a997",
            "3deb996fa6434be78a6c8e1b1242cd88",
            "cef44826a2954f4aa93a679eb9e679ca",
            "2c1f524677d34ab4893a4328abf6673d",
            "8ff5b5ac6fd543ea8ac104bdfd1b4389",
            "63a50945615c46009f6fb0b126cecd99",
            "f82430d8467745d791ca7f794f9395e1",
            "ffeaebb887594ebbb13c5c27c49e4b87",
            "ce8225da4ff14deb9148d70fb2e779d9",
            "fe2fe07615a34a258d5404776fd979a7",
            "e01b19af908f43a3b3d4453a70052a54",
            "d0ad64357ac2449cae63c9c6dd9b2063",
            "fbcd307210e646688b156b1673d07d04",
            "b7a91bbfbc8145e0942f6b6d44a0b563",
            "fcc58354a4e74829b93c328d257afcee",
            "e18635a6debf4bb8b5e4633eeb652714",
            "bd3ecc735d1d4a59946cb4807ba7bd9a",
            "052659564d544b868ca5f1630ce7d794",
            "005f3b6dd23b454f855b7eb35b9b9845",
            "dd1a4e2718ba4adeaf1432a8634ab50c",
            "e82836e4c6624639a177dc81a4e1f13c",
            "84d1ed1c8f064ca78526f9cb331f796e",
            "9d1ade6fd4714284ad734656072ddf85",
            "1c645ea089524c42a9b95832541f7b20",
            "9b3ffcb4ecbd4272bac71e8e907910d2",
            "ed421f80bd89455496291da3041b255e",
            "c18dce84c0214a58bb7d3da26af3797d",
            "77310020500d48518cd860c934a5fcbe",
            "dee1141f0d9f4fddaab220835b28ebf1",
            "b0a3396b2947431a9e51ad0bf9388759",
            "787014b746584fd6816caab602ba83e5",
            "b0b011147e844090bd2468d11ef20fb5",
            "59be7a96a0224c3eacff2727a2315804",
            "1118dd0dd78f4eeda72abfadb29637f9",
            "b0f00ee547314dec96eda13950de120f",
            "c75646c5fab64b2b88cba270c0654d2b",
            "6e4e2753b4c44c899b2ed61c7190724d"
          ]
        },
        "id": "4yjyhkT208_F",
        "outputId": "0ec490ab-be2c-4c78-9339-3199936847ec"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Using 16bit native Automatic Mixed Precision (AMP)\n",
            "GPU available: True (cuda), used: True\n",
            "TPU available: False, using: 0 TPU cores\n",
            "IPU available: False, using: 0 IPUs\n",
            "HPU available: False, using: 0 HPUs\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Positive Weights for the loss are -> tensor([1.], device='cuda:0', dtype=torch.float64)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torchmetrics/utilities/prints.py:36: UserWarning: Metric `AveragePrecision` will save all targets and predictions in buffer. For large datasets this may lead to large memory footprint.\n",
            "  warnings.warn(*args, **kwargs)\n",
            "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
            "\n",
            "  | Name       | Type             | Params\n",
            "------------------------------------------------\n",
            "0 | backbone   | RegNet           | 39.6 M\n",
            "1 | classifier | Sequential       | 2.1 M \n",
            "2 | avg_prc    | AveragePrecision | 0     \n",
            "3 | recall     | Recall           | 0     \n",
            "4 | prec       | Precision        | 0     \n",
            "------------------------------------------------\n",
            "4.1 M     Trainable params\n",
            "37.7 M    Non-trainable params\n",
            "41.7 M    Total params\n",
            "83.439    Total estimated model params size (MB)\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Sanity Checking: 0it [00:00, ?it/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "35272b94d6bd46509e6fe9cc82280469"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Training: 0it [00:00, ?it/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "8ff5b5ac6fd543ea8ac104bdfd1b4389"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Validation: 0it [00:00, ?it/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "e18635a6debf4bb8b5e4633eeb652714"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/pytorch_lightning/trainer/trainer.py:653: UserWarning: Detected KeyboardInterrupt, attempting graceful shutdown...\n",
            "  rank_zero_warn(\"Detected KeyboardInterrupt, attempting graceful shutdown...\")\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Positive Weights for the loss are -> tensor([1.], dtype=torch.float64)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Predicting: 90it [00:00, ?it/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "c18dce84c0214a58bb7d3da26af3797d"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:70: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray.\n"
          ]
        }
      ],
      "source": [
        "batch_size = 128\n",
        "simclr_model, pred_df = train_simclr(\n",
        "                            model_name = 'Effusion_2048_hidden_full',\n",
        "                              batch_size=batch_size ,\n",
        "                            hidden_dim=2048,\n",
        "                            lr= 5e-4 ,\n",
        "                            max_epochs = 4\n",
        "                            )"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def add_metrics(df):\n",
        "  df['tn'] = (df['bin_pred'] == 0) &  (df['Effusion'] == 0)\n",
        "  df['fn'] = (df['bin_pred'] == 0) &  (df['Effusion'] == 1)\n",
        "  df['fp'] = (df['bin_pred'] == 1) &  (df['Effusion'] == 0)\n",
        "  df['tp'] = (df['bin_pred'] == 1) &  (df['Effusion'] == 1)\n",
        "  return df"
      ],
      "metadata": {
        "id": "SXD31bVMmCJj"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "THRESHOLD = .5\n",
        "pred_df['bin_pred'] = (pred_df['predictions'] >  THRESHOLD).astype('float')\n",
        "working_df = pred_df.loc[:, ['image_id', 'position', 'Effusion', 'bin_pred']]\n",
        "working_df = add_metrics(working_df)\n",
        "g_position = working_df.groupby('position')\n",
        "g_sum = g_position.sum()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8mbq5eX1nv2j",
        "outputId": "1bc510cc-ee89-48ee-bdb4-2ecae5b57df1"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Effusion    1017.0\n",
              "bin_pred     760.0\n",
              "tn          1014.0\n",
              "fn           452.0\n",
              "fp           195.0\n",
              "tp           565.0\n",
              "dtype: float64"
            ]
          },
          "metadata": {},
          "execution_count": 50
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BJRQZVRAscxF"
      },
      "source": [
        "how should we approach this?\n",
        "# Error analysis\n",
        "\n",
        "I have dhe Data ahd the model and now i rant to sxtRaxt the error profi\n",
        "le Out of itrsT \n",
        "\n",
        "I should probably make a df inside only with all the Predictions attached iN it\n",
        "and after that use that to understaNd the error"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "KISqmPdgiA0S"
      },
      "outputs": [],
      "source": [
        "model = simclr_model\n",
        "model.requires_grad_(True)\n",
        "features_conv = nn.Sequential(\n",
        "    model.backbone.stem,\n",
        "    model.backbone.trunk_output\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 321
        },
        "id": "eqHvplTgCesM",
        "outputId": "e4047d38-f51b-4dd4-fa9a-6786f685d3c9"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "torch.Size([1, 1920, 9, 9])\n",
            "torch.Size([1, 1920])\n",
            "tensor([[0.5118]], grad_fn=<AddmmBackward0>) tensor([1.], dtype=torch.float64)\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "class Regnet_Grad(nn.Module):\n",
        "  def __init__(self,):\n",
        "    super().__init__()\n",
        "    self.backbone = model.backbone\n",
        "    self.features_conv = features_conv\n",
        "    self.cls = model.classifier\n",
        "    self.pretrained_head = model.backbone.fc\n",
        "    # self.max_pool = nn.MaxPool2d(kernel_size = 4, stride = 4, padding =0, dilation = 2, ceil_mode = False)\n",
        "    self.max_pool = model.backbone.avgpool\n",
        "    self.gradients = None\n",
        "\n",
        "  def activations_hook(self, grad):\n",
        "    self.gradients = grad\n",
        "\n",
        "  def forward(self, x):\n",
        "    self.zero_grad()\n",
        "    x = self.features_conv(x)\n",
        "    # x.requires_grad = True\n",
        "    print(x.shape)\n",
        "    h = x.register_hook(self.activations_hook)\n",
        "\n",
        "    x = self.max_pool(x)\n",
        "    x = x.flatten(1,-1)\n",
        "    print(x.shape)\n",
        "    x = self.pretrained_head(x)\n",
        "    x = self.cls(x)\n",
        "    return x\n",
        "\n",
        "  def get_activations_gradients(self):\n",
        "    return self.gradients\n",
        "\n",
        "  def get_activations(self, x):\n",
        "    return self.features_conv(x)\n",
        "\n",
        "grad_model = Regnet_Grad()\n",
        "ds= ClassificationDataset(final_df)\n",
        "dl = DataLoader(ds, 1, shuffle = True)\n",
        "imgs, labels = next(iter(dl))\n",
        "preds=  grad_model(imgs)\n",
        "# preds += 1.\n",
        "preds.backward()\n",
        "from torchvision.transforms.functional import InterpolationMode\n",
        "ds= ClassificationDataset(final_df)\n",
        "dl = DataLoader(ds, 1, shuffle = True)\n",
        "imgs, labels = next(iter(dl))\n",
        "plt.set_cmap('gray')\n",
        "gradients = grad_model.get_activations_gradients()\n",
        "# Somehting odd happened with the registed hook as teh shape here is 1920 x 9 x9 which is same as the features conv\n",
        "pooled_gradients = torch.mean(gradients, dim = [0,2,3])\n",
        "activations = grad_model.get_activations(imgs).detach()\n",
        "\n",
        "pooled_gradients.shape, activations.shape\n",
        "for i in range(pooled_gradients.shape[0]):\n",
        "  activations[:,i,:,:] *= pooled_gradients[i]\n",
        "\n",
        "heatmap =torch.mean(activations, dim = 1).squeeze()\n",
        "heatmap = F.relu(heatmap)\n",
        "heatmap /= torch.max(heatmap)\n",
        "# plt.imshow(heatmap)\n",
        "# print(heatmap.shape)\n",
        "heatmap_up = T.Resize((284,284),interpolation = InterpolationMode.BICUBIC)(heatmap[None,:])\n",
        "plt.imshow(imgs[0,0,:])\n",
        "plt.imshow(heatmap_up[0], cmap = plt.cm.jet, alpha = 0.4)\n",
        "print(preds, labels)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bQEUykcYY1r4",
        "outputId": "07b136d8-61a7-4627-9d28-ac92ead618ba"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "Classifier(\n",
              "  (backbone): RegNet(\n",
              "    (stem): Sequential(\n",
              "      (0): Conv2d(1, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
              "      (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "      (2): ReLU(inplace=True)\n",
              "    )\n",
              "    (trunk_output): Sequential(\n",
              "      (block1): AnyStage(\n",
              "        (block1-0): ResBottleneckBlock(\n",
              "          (proj): Conv2dNormActivation(\n",
              "            (0): Conv2d(32, 80, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
              "            (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "          )\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(32, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(80, 80, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(80, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block1-1): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(80, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(80, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(80, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "      )\n",
              "      (block2): AnyStage(\n",
              "        (block2-0): ResBottleneckBlock(\n",
              "          (proj): Conv2dNormActivation(\n",
              "            (0): Conv2d(80, 240, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
              "            (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "          )\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(80, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=2, bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block2-1): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block2-2): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block2-3): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block2-4): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "      )\n",
              "      (block3): AnyStage(\n",
              "        (block3-0): ResBottleneckBlock(\n",
              "          (proj): Conv2dNormActivation(\n",
              "            (0): Conv2d(240, 720, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
              "            (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "          )\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(240, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-1): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-2): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-3): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-4): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-5): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-6): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-7): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-8): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-9): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-10): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-11): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-12): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-13): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "        (block3-14): ResBottleneckBlock(\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=6, bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 720, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(720, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "      )\n",
              "      (block4): AnyStage(\n",
              "        (block4-0): ResBottleneckBlock(\n",
              "          (proj): Conv2dNormActivation(\n",
              "            (0): Conv2d(720, 1920, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
              "            (1): BatchNorm2d(1920, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "          )\n",
              "          (f): BottleneckTransform(\n",
              "            (a): Conv2dNormActivation(\n",
              "              (0): Conv2d(720, 1920, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(1920, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (b): Conv2dNormActivation(\n",
              "              (0): Conv2d(1920, 1920, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=16, bias=False)\n",
              "              (1): BatchNorm2d(1920, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "              (2): ReLU(inplace=True)\n",
              "            )\n",
              "            (c): Conv2dNormActivation(\n",
              "              (0): Conv2d(1920, 1920, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
              "              (1): BatchNorm2d(1920, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
              "            )\n",
              "          )\n",
              "          (activation): ReLU(inplace=True)\n",
              "        )\n",
              "      )\n",
              "    )\n",
              "    (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n",
              "    (fc): Linear(in_features=1920, out_features=1024, bias=True)\n",
              "  )\n",
              "  (classifier): Sequential(\n",
              "    (0): Dropout(p=0.1, inplace=False)\n",
              "    (1): ReLU(inplace=True)\n",
              "    (2): Linear(in_features=1024, out_features=1024, bias=True)\n",
              "    (3): Dropout(p=0.1, inplace=False)\n",
              "    (4): ReLU(inplace=True)\n",
              "    (5): Linear(in_features=1024, out_features=1, bias=True)\n",
              "  )\n",
              "  (avg_prc): AveragePrecision()\n",
              ")"
            ]
          },
          "execution_count": 15,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "simclr_model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 432
        },
        "id": "cib6sr-PXrL1",
        "outputId": "e17dae0f-fb32-4daa-d65a-ad7be34f123f"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "torch.Size([16, 1, 284, 284])\n"
          ]
        },
        {
          "ename": "AxisError",
          "evalue": "ignored",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mAxisError\u001b[0m                                 Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-27-d1ac145a1da2>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     24\u001b[0m \u001b[0mcam\u001b[0m\u001b[0;34m=\u001b[0m \u001b[0mGradCAM\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msimclr_model\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     25\u001b[0m \u001b[0;31m# targets = [ClassifierOutputTarget(labels)]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 26\u001b[0;31m \u001b[0mgray_cam\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcam\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mimgs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     27\u001b[0m \u001b[0mfig\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0max\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mplt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msubplots\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfigsize\u001b[0m\u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m30\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m30\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     28\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/pytorch_grad_cam/base_cam.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, input_tensor, targets, aug_smooth, eigen_smooth)\u001b[0m\n\u001b[1;32m    187\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    188\u001b[0m         return self.forward(input_tensor,\n\u001b[0;32m--> 189\u001b[0;31m                             targets, eigen_smooth)\n\u001b[0m\u001b[1;32m    190\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    191\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__del__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/pytorch_grad_cam/base_cam.py\u001b[0m in \u001b[0;36mforward\u001b[0;34m(self, input_tensor, targets, eigen_smooth)\u001b[0m\n\u001b[1;32m     95\u001b[0m         cam_per_layer = self.compute_cam_per_layer(input_tensor,\n\u001b[1;32m     96\u001b[0m                                                    \u001b[0mtargets\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 97\u001b[0;31m                                                    eigen_smooth)\n\u001b[0m\u001b[1;32m     98\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0maggregate_multi_layers\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcam_per_layer\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     99\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/pytorch_grad_cam/base_cam.py\u001b[0m in \u001b[0;36mcompute_cam_per_layer\u001b[0;34m(self, input_tensor, targets, eigen_smooth)\u001b[0m\n\u001b[1;32m    130\u001b[0m                                      \u001b[0mlayer_activations\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    131\u001b[0m                                      \u001b[0mlayer_grads\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 132\u001b[0;31m                                      eigen_smooth)\n\u001b[0m\u001b[1;32m    133\u001b[0m             \u001b[0mcam\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmaximum\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcam\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    134\u001b[0m             \u001b[0mscaled\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mscale_cam_image\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcam\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget_size\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/pytorch_grad_cam/base_cam.py\u001b[0m in \u001b[0;36mget_cam_image\u001b[0;34m(self, input_tensor, target_layer, targets, activations, grads, eigen_smooth)\u001b[0m\n\u001b[1;32m     52\u001b[0m                                        \u001b[0mtargets\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     53\u001b[0m                                        \u001b[0mactivations\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 54\u001b[0;31m                                        grads)\n\u001b[0m\u001b[1;32m     55\u001b[0m         \u001b[0mweighted_activations\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mweights\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0mactivations\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     56\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0meigen_smooth\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/pytorch_grad_cam/grad_cam.py\u001b[0m in \u001b[0;36mget_cam_weights\u001b[0;34m(self, input_tensor, target_layer, target_category, activations, grads)\u001b[0m\n\u001b[1;32m     20\u001b[0m                         \u001b[0mactivations\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     21\u001b[0m                         grads):\n\u001b[0;32m---> 22\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmean\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mgrads\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m3\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m<__array_function__ internals>\u001b[0m in \u001b[0;36mmean\u001b[0;34m(*args, **kwargs)\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/numpy/core/fromnumeric.py\u001b[0m in \u001b[0;36mmean\u001b[0;34m(a, axis, dtype, out, keepdims, where)\u001b[0m\n\u001b[1;32m   3439\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3440\u001b[0m     return _methods._mean(a, axis=axis, dtype=dtype,\n\u001b[0;32m-> 3441\u001b[0;31m                           out=out, **kwargs)\n\u001b[0m\u001b[1;32m   3442\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3443\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/numpy/core/_methods.py\u001b[0m in \u001b[0;36m_mean\u001b[0;34m(a, axis, dtype, out, keepdims, where)\u001b[0m\n\u001b[1;32m    165\u001b[0m     \u001b[0mis_float16_result\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    166\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 167\u001b[0;31m     \u001b[0mrcount\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_count_reduce_items\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0marr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mkeepdims\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mkeepdims\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mwhere\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mwhere\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    168\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mrcount\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mwhere\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mTrue\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0mumr_any\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrcount\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mNone\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    169\u001b[0m         \u001b[0mwarnings\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwarn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"Mean of empty slice.\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mRuntimeWarning\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mstacklevel\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/numpy/core/_methods.py\u001b[0m in \u001b[0;36m_count_reduce_items\u001b[0;34m(arr, axis, keepdims, where)\u001b[0m\n\u001b[1;32m     74\u001b[0m         \u001b[0mitems\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mintp\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     75\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0max\u001b[0m \u001b[0;32min\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 76\u001b[0;31m             \u001b[0mitems\u001b[0m \u001b[0;34m*=\u001b[0m \u001b[0marr\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mmu\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnormalize_axis_index\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0max\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0marr\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mndim\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     77\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     78\u001b[0m         \u001b[0;31m# TODO: Optimize case when `where` is broadcast along a non-reduction\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mAxisError\u001b[0m: axis 2 is out of bounds for array of dimension 0"
          ]
        }
      ],
      "source": [
        "from pytorch_grad_cam import GradCAM, HiResCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, FullGrad\n",
        "from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget\n",
        "from pytorch_grad_cam.utils.image import show_cam_on_image\n",
        "# simclr_model.cpu()\n",
        "trunk = simclr_model.backbone.trunk_output\n",
        "target = [\n",
        "    trunk.get_submodule('block2').get_submodule('block2-2').f.c[0],\n",
        "    # trunk.get_submodule('block2').get_submodule('block2-1').f.c[0],\n",
        "    # trunk.get_submodule('block2').get_submodule('block2-3').f.c[0],\n",
        "    # trunk.get_submodule('block2').get_submodule('block2-4').f.c[0],\n",
        "    # trunk.get_submodule('block2').get_submodule('block2-5').f.c[0],\n",
        "    # trunk.get_submodule('block2').get_submodule('block2-6').f.c[0],\n",
        "    # trunk.get_submodule('block2').get_submodule('block2-7').f.c[0],\n",
        "    # trunk.get_submodule('block1').get_submodule('block1-1').f.c[0],\n",
        "    # trunk.get_submodule('block3').get_submodule('block3-3').f.c[0],\n",
        "]\n",
        "# target = [model.encoder.trunk_output.block2.get_submodule('block2-2').f.c[0]]\n",
        "\n",
        "ds = ClassificationDataset(final_df)\n",
        "dl = DataLoader(ds, 16, shuffle = True)\n",
        "imgs, labels= next(iter(dl))\n",
        "imgs = imgs.cuda()\n",
        "print(imgs.shape)\n",
        "cam= GradCAM(simclr_model, target, True)\n",
        "# targets = [ClassifierOutputTarget(labels)]\n",
        "gray_cam = cam(imgs)\n",
        "fig, ax = plt.subplots(4,4, figsize= (30,30))\n",
        "\n",
        "ax = ax.ravel()\n",
        "for i in range(16):\n",
        "  ax[i].imshow(imgs[i,0,:], cmap = 'gray')\n",
        "  ax[i].imshow(gray_cam[i], alpha = .5, cmap = 'jet')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3K4Yr71AAugk"
      },
      "source": [
        "### Implemet a good expandable pipeline for testing purposes which can be used in future projects also.T"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "caDRwtT-AubT"
      },
      "source": []
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [
        "7-byMfPGgEFe",
        "P6tqstiJo983",
        "mRMDIrcWngrw"
      ],
      "machine_shape": "hm",
      "provenance": [],
      "mount_file_id": "19s7nE08QXyFleWDpooOAcuwM0cD8f_Do",
      "authorship_tag": "ABX9TyNPshWfQMFidfUUDLpVfCZI",
      "include_colab_link": true
    },
    "gpuClass": "standard",
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "35272b94d6bd46509e6fe9cc82280469": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_8205ae282d8747f1bfa12729ab908e27",
              "IPY_MODEL_f2cde6a705624a87a344e1adb92be88d",
              "IPY_MODEL_13252b4efb704551a383696b7251beed"
            ],
            "layout": "IPY_MODEL_aa10847a99f74929b601c6252747d86f"
          }
        },
        "8205ae282d8747f1bfa12729ab908e27": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_2709658a7ce9477fa78c5ffbfa922178",
            "placeholder": "​",
            "style": "IPY_MODEL_998217366fdc40adaa7aa014ce70754c",
            "value": "Sanity Checking DataLoader 0: 100%"
          }
        },
        "f2cde6a705624a87a344e1adb92be88d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_83bda3c98e2f491b935ce42904a4a997",
            "max": 2,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_3deb996fa6434be78a6c8e1b1242cd88",
            "value": 2
          }
        },
        "13252b4efb704551a383696b7251beed": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_cef44826a2954f4aa93a679eb9e679ca",
            "placeholder": "​",
            "style": "IPY_MODEL_2c1f524677d34ab4893a4328abf6673d",
            "value": " 2/2 [00:00&lt;00:00,  6.73it/s]"
          }
        },
        "aa10847a99f74929b601c6252747d86f": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": "inline-flex",
            "flex": null,
            "flex_flow": "row wrap",
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": "100%"
          }
        },
        "2709658a7ce9477fa78c5ffbfa922178": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "998217366fdc40adaa7aa014ce70754c": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "83bda3c98e2f491b935ce42904a4a997": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": "2",
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "3deb996fa6434be78a6c8e1b1242cd88": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "cef44826a2954f4aa93a679eb9e679ca": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "2c1f524677d34ab4893a4328abf6673d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "8ff5b5ac6fd543ea8ac104bdfd1b4389": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_63a50945615c46009f6fb0b126cecd99",
              "IPY_MODEL_f82430d8467745d791ca7f794f9395e1",
              "IPY_MODEL_ffeaebb887594ebbb13c5c27c49e4b87"
            ],
            "layout": "IPY_MODEL_ce8225da4ff14deb9148d70fb2e779d9"
          }
        },
        "63a50945615c46009f6fb0b126cecd99": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_fe2fe07615a34a258d5404776fd979a7",
            "placeholder": "​",
            "style": "IPY_MODEL_e01b19af908f43a3b3d4453a70052a54",
            "value": "Epoch 1:  38%"
          }
        },
        "f82430d8467745d791ca7f794f9395e1": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_d0ad64357ac2449cae63c9c6dd9b2063",
            "max": 208,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_fbcd307210e646688b156b1673d07d04",
            "value": 80
          }
        },
        "ffeaebb887594ebbb13c5c27c49e4b87": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_b7a91bbfbc8145e0942f6b6d44a0b563",
            "placeholder": "​",
            "style": "IPY_MODEL_fcc58354a4e74829b93c328d257afcee",
            "value": " 80/208 [00:13&lt;00:21,  5.90it/s, loss=0.603, v_num=10, train_prc=0.713, train_precision=0.680, train_recall=0.548, train_loss=0.599, val_prc=0.728, val_precision=0.707, val_recall=0.650, val_loss=0.612]"
          }
        },
        "ce8225da4ff14deb9148d70fb2e779d9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": "inline-flex",
            "flex": null,
            "flex_flow": "row wrap",
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": "100%"
          }
        },
        "fe2fe07615a34a258d5404776fd979a7": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "e01b19af908f43a3b3d4453a70052a54": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "d0ad64357ac2449cae63c9c6dd9b2063": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": "2",
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "fbcd307210e646688b156b1673d07d04": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "b7a91bbfbc8145e0942f6b6d44a0b563": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "fcc58354a4e74829b93c328d257afcee": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "e18635a6debf4bb8b5e4633eeb652714": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_bd3ecc735d1d4a59946cb4807ba7bd9a",
              "IPY_MODEL_052659564d544b868ca5f1630ce7d794",
              "IPY_MODEL_005f3b6dd23b454f855b7eb35b9b9845"
            ],
            "layout": "IPY_MODEL_dd1a4e2718ba4adeaf1432a8634ab50c"
          }
        },
        "bd3ecc735d1d4a59946cb4807ba7bd9a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_e82836e4c6624639a177dc81a4e1f13c",
            "placeholder": "​",
            "style": "IPY_MODEL_84d1ed1c8f064ca78526f9cb331f796e",
            "value": "Validation DataLoader 0: 100%"
          }
        },
        "052659564d544b868ca5f1630ce7d794": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_9d1ade6fd4714284ad734656072ddf85",
            "max": 32,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_1c645ea089524c42a9b95832541f7b20",
            "value": 32
          }
        },
        "005f3b6dd23b454f855b7eb35b9b9845": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_9b3ffcb4ecbd4272bac71e8e907910d2",
            "placeholder": "​",
            "style": "IPY_MODEL_ed421f80bd89455496291da3041b255e",
            "value": " 32/32 [00:04&lt;00:00,  7.51it/s]"
          }
        },
        "dd1a4e2718ba4adeaf1432a8634ab50c": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": "inline-flex",
            "flex": null,
            "flex_flow": "row wrap",
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": "100%"
          }
        },
        "e82836e4c6624639a177dc81a4e1f13c": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "84d1ed1c8f064ca78526f9cb331f796e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "9d1ade6fd4714284ad734656072ddf85": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": "2",
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "1c645ea089524c42a9b95832541f7b20": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "9b3ffcb4ecbd4272bac71e8e907910d2": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ed421f80bd89455496291da3041b255e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "c18dce84c0214a58bb7d3da26af3797d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_77310020500d48518cd860c934a5fcbe",
              "IPY_MODEL_dee1141f0d9f4fddaab220835b28ebf1",
              "IPY_MODEL_b0a3396b2947431a9e51ad0bf9388759"
            ],
            "layout": "IPY_MODEL_787014b746584fd6816caab602ba83e5"
          }
        },
        "77310020500d48518cd860c934a5fcbe": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_b0b011147e844090bd2468d11ef20fb5",
            "placeholder": "​",
            "style": "IPY_MODEL_59be7a96a0224c3eacff2727a2315804",
            "value": "Predicting DataLoader 0: 100%"
          }
        },
        "dee1141f0d9f4fddaab220835b28ebf1": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_1118dd0dd78f4eeda72abfadb29637f9",
            "max": 32,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_b0f00ee547314dec96eda13950de120f",
            "value": 32
          }
        },
        "b0a3396b2947431a9e51ad0bf9388759": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_c75646c5fab64b2b88cba270c0654d2b",
            "placeholder": "​",
            "style": "IPY_MODEL_6e4e2753b4c44c899b2ed61c7190724d",
            "value": " 32/32 [00:04&lt;00:00,  7.56it/s]"
          }
        },
        "787014b746584fd6816caab602ba83e5": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": "inline-flex",
            "flex": null,
            "flex_flow": "row wrap",
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": "100%"
          }
        },
        "b0b011147e844090bd2468d11ef20fb5": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "59be7a96a0224c3eacff2727a2315804": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "1118dd0dd78f4eeda72abfadb29637f9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": "2",
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "b0f00ee547314dec96eda13950de120f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "c75646c5fab64b2b88cba270c0654d2b": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "6e4e2753b4c44c899b2ed61c7190724d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        }
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}