nih-vit-gradcam.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "7e7ffaa2",
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5",
    "execution": {
     "iopub.execute_input": "2024-01-07T19:04:43.965818Z",
     "iopub.status.busy": "2024-01-07T19:04:43.965121Z",
     "iopub.status.idle": "2024-01-07T19:04:57.450600Z",
     "shell.execute_reply": "2024-01-07T19:04:57.449740Z"
    },
    "papermill": {
     "duration": 13.498307,
     "end_time": "2024-01-07T19:04:57.453060",
     "exception": false,
     "start_time": "2024-01-07T19:04:43.954753",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.10/site-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.16.5 and <1.23.0 is required for this version of SciPy (detected version 1.24.3\n",
      "  warnings.warn(f\"A NumPy version >={np_minversion} and <{np_maxversion}\"\n",
      "/opt/conda/lib/python3.10/site-packages/tensorflow_addons/utils/tfa_eol_msg.py:23: UserWarning: \n",
      "\n",
      "TensorFlow Addons (TFA) has ended development and introduction of new features.\n",
      "TFA has entered a minimal maintenance and release mode until a planned end of life in May 2024.\n",
      "Please modify downstream libraries to take dependencies from other repositories in our TensorFlow community (e.g. Keras, Keras-CV, and Keras-NLP). \n",
      "\n",
      "For more information see: https://github.com/tensorflow/addons/issues/2807 \n",
      "\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "import numpy as np \n",
    "import pandas as pd\n",
    "import os\n",
    "import glob\n",
    "import itertools\n",
    "import time\n",
    "import seaborn as sns\n",
    "import tensorflow as tf\n",
    "import tensorflow_addons as tfa\n",
    "import matplotlib.pyplot as plt\n",
    "from tensorflow.keras import backend as K\n",
    "from tensorflow.keras.applications import EfficientNetB0\n",
    "from tensorflow.keras.models import Sequential, Model\n",
    "from tensorflow.keras.layers import LayerNormalization, Input, LSTM, GRU, TimeDistributed\n",
    "from tensorflow.keras.layers import Conv2D, BatchNormalization, ReLU, MaxPooling2D, Flatten, Dropout, GlobalAveragePooling2D, Dense, Softmax, Bidirectional, GlobalAveragePooling1D\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from pandas import DataFrame\n",
    "from typing import List\n",
    "from PIL import Image\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import roc_auc_score, roc_curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "5c8f2345",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:04:57.472639Z",
     "iopub.status.busy": "2024-01-07T19:04:57.472076Z",
     "iopub.status.idle": "2024-01-07T19:05:12.028845Z",
     "shell.execute_reply": "2024-01-07T19:05:12.027912Z"
    },
    "papermill": {
     "duration": 14.568676,
     "end_time": "2024-01-07T19:05:12.030918",
     "exception": false,
     "start_time": "2024-01-07T19:04:57.462242",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Image Index</th>\n",
       "      <th>Finding Labels</th>\n",
       "      <th>Follow-up #</th>\n",
       "      <th>Patient ID</th>\n",
       "      <th>Patient Age</th>\n",
       "      <th>Patient Gender</th>\n",
       "      <th>View Position</th>\n",
       "      <th>OriginalImageWidth</th>\n",
       "      <th>OriginalImageHeight</th>\n",
       "      <th>OriginalImagePixelSpacingX</th>\n",
       "      <th>...</th>\n",
       "      <th>Edema</th>\n",
       "      <th>Cardiomegaly</th>\n",
       "      <th>Mass</th>\n",
       "      <th>Pneumothorax</th>\n",
       "      <th>Consolidation</th>\n",
       "      <th>Fibrosis</th>\n",
       "      <th>Pleural_Thickening</th>\n",
       "      <th>Emphysema</th>\n",
       "      <th>Nodule</th>\n",
       "      <th>Infiltration</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>00000001_000.png</td>\n",
       "      <td>Cardiomegaly</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>58</td>\n",
       "      <td>M</td>\n",
       "      <td>PA</td>\n",
       "      <td>2682</td>\n",
       "      <td>2749</td>\n",
       "      <td>0.143</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>00000001_001.png</td>\n",
       "      <td>Cardiomegaly|Emphysema</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>58</td>\n",
       "      <td>M</td>\n",
       "      <td>PA</td>\n",
       "      <td>2894</td>\n",
       "      <td>2729</td>\n",
       "      <td>0.143</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>00000001_002.png</td>\n",
       "      <td>Cardiomegaly|Effusion</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>58</td>\n",
       "      <td>M</td>\n",
       "      <td>PA</td>\n",
       "      <td>2500</td>\n",
       "      <td>2048</td>\n",
       "      <td>0.168</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>00000002_000.png</td>\n",
       "      <td>No Finding</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>81</td>\n",
       "      <td>M</td>\n",
       "      <td>PA</td>\n",
       "      <td>2500</td>\n",
       "      <td>2048</td>\n",
       "      <td>0.171</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>00000003_000.png</td>\n",
       "      <td>Hernia</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>81</td>\n",
       "      <td>F</td>\n",
       "      <td>PA</td>\n",
       "      <td>2582</td>\n",
       "      <td>2991</td>\n",
       "      <td>0.143</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 27 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        Image Index          Finding Labels  Follow-up #  Patient ID  \\\n",
       "0  00000001_000.png            Cardiomegaly            0           1   \n",
       "1  00000001_001.png  Cardiomegaly|Emphysema            1           1   \n",
       "2  00000001_002.png   Cardiomegaly|Effusion            2           1   \n",
       "3  00000002_000.png              No Finding            0           2   \n",
       "4  00000003_000.png                  Hernia            0           3   \n",
       "\n",
       "   Patient Age Patient Gender View Position  OriginalImageWidth  \\\n",
       "0           58              M            PA                2682   \n",
       "1           58              M            PA                2894   \n",
       "2           58              M            PA                2500   \n",
       "3           81              M            PA                2500   \n",
       "4           81              F            PA                2582   \n",
       "\n",
       "   OriginalImageHeight  OriginalImagePixelSpacingX  ...  Edema  Cardiomegaly  \\\n",
       "0                 2749                       0.143  ...    0.0           1.0   \n",
       "1                 2729                       0.143  ...    0.0           1.0   \n",
       "2                 2048                       0.168  ...    0.0           1.0   \n",
       "3                 2048                       0.171  ...    0.0           0.0   \n",
       "4                 2991                       0.143  ...    0.0           0.0   \n",
       "\n",
       "   Mass  Pneumothorax  Consolidation  Fibrosis  Pleural_Thickening  Emphysema  \\\n",
       "0   0.0           0.0            0.0       0.0                 0.0        0.0   \n",
       "1   0.0           0.0            0.0       0.0                 0.0        1.0   \n",
       "2   0.0           0.0            0.0       0.0                 0.0        0.0   \n",
       "3   0.0           0.0            0.0       0.0                 0.0        0.0   \n",
       "4   0.0           0.0            0.0       0.0                 0.0        0.0   \n",
       "\n",
       "   Nodule  Infiltration  \n",
       "0     0.0           0.0  \n",
       "1     0.0           0.0  \n",
       "2     0.0           0.0  \n",
       "3     0.0           0.0  \n",
       "4     0.0           0.0  \n",
       "\n",
       "[5 rows x 27 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Read the dataframe\n",
    "df = pd.read_csv('/kaggle/input/data/Data_Entry_2017.csv')\n",
    "\n",
    "# Rename the columns\n",
    "df = df.rename(columns={\n",
    "    \"OriginalImagePixelSpacing[x\": \"OriginalImagePixelSpacingX\",\n",
    "    \"y]\": \"OriginalImagePixelSpacingY\",\n",
    "    \"OriginalImage[Width\": \"OriginalImageWidth\",\n",
    "    \"Height]\": \"OriginalImageHeight\"\n",
    "})\n",
    "### Format the label by a one hot encoding\n",
    "# Get all the label\n",
    "unique_labels = set(itertools.chain.from_iterable(df[\"Finding Labels\"].apply(lambda x : x.split('|')).values)) \n",
    "\n",
    "# Initialize a empty matrix \n",
    "#one_hot_labels = pd.DataFrame(0.0, index=np.arange(len(df)), columns=unique_labels)\n",
    "one_hot_labels = pd.DataFrame(0.0, index=np.arange(len(df)), columns=list(unique_labels))\n",
    "\n",
    "# For each row, we get the associated labels and set a 1 to the new corresponding column label \n",
    "for index, row in df.iterrows():\n",
    "    labels = row[\"Finding Labels\"].split('|')\n",
    "    for label in labels:\n",
    "        one_hot_labels.iloc[index][label] = 1.0\n",
    "\n",
    "# Then, we concatane this new dataframe to our original data\n",
    "df = pd.concat([df, one_hot_labels], axis=1)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "946a28ea",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.050844Z",
     "iopub.status.busy": "2024-01-07T19:05:12.050132Z",
     "iopub.status.idle": "2024-01-07T19:05:12.102547Z",
     "shell.execute_reply": "2024-01-07T19:05:12.101474Z"
    },
    "papermill": {
     "duration": 0.064743,
     "end_time": "2024-01-07T19:05:12.104741",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.039998",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of training set : 75120  - testing set: 37000\n"
     ]
    }
   ],
   "source": [
    "df_train, df_test = train_test_split(df, test_size = 0.33, random_state = 42)\n",
    "print(\"Size of training set :\", len(df_train), \" - testing set:\", len(df_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "6c3e4087",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.124077Z",
     "iopub.status.busy": "2024-01-07T19:05:12.123727Z",
     "iopub.status.idle": "2024-01-07T19:05:12.153665Z",
     "shell.execute_reply": "2024-01-07T19:05:12.152695Z"
    },
    "papermill": {
     "duration": 0.042067,
     "end_time": "2024-01-07T19:05:12.155700",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.113633",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9769\n"
     ]
    }
   ],
   "source": [
    "patient_train = set(df_train[\"Patient ID\"].values)\n",
    "patient_test  = set(df_test[\"Patient ID\"].values)\n",
    "\n",
    "leakage_patient = patient_train.intersection(patient_test)\n",
    "print(len(leakage_patient))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "de3789b8",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.174958Z",
     "iopub.status.busy": "2024-01-07T19:05:12.174591Z",
     "iopub.status.idle": "2024-01-07T19:05:12.199316Z",
     "shell.execute_reply": "2024-01-07T19:05:12.198257Z"
    },
    "papermill": {
     "duration": 0.036857,
     "end_time": "2024-01-07T19:05:12.201466",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.164609",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train\n",
      "0.2672790202342918\n",
      "75120\n",
      "20078\n",
      "\n",
      "Test\n",
      "0.18627027027027027\n",
      "37000\n",
      "6892\n"
     ]
    }
   ],
   "source": [
    "# How much this represent based on our set\n",
    "train_without_leakage = df_train[~df_train[\"Patient ID\"].isin(leakage_patient)]\n",
    "\n",
    "print(\"Train\")\n",
    "print(len(train_without_leakage) / len(df_train))\n",
    "print(len(df_train))\n",
    "print(len(train_without_leakage))\n",
    "\n",
    "print()\n",
    "print(\"Test\")\n",
    "test_without_leakage = df_test[~df_test[\"Patient ID\"].isin(leakage_patient)]\n",
    "print(len(test_without_leakage) / len(df_test))\n",
    "print(len(df_test))\n",
    "print(len(test_without_leakage))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "63eeae3d",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.221133Z",
     "iopub.status.busy": "2024-01-07T19:05:12.220802Z",
     "iopub.status.idle": "2024-01-07T19:05:12.225800Z",
     "shell.execute_reply": "2024-01-07T19:05:12.224850Z"
    },
    "papermill": {
     "duration": 0.017141,
     "end_time": "2024-01-07T19:05:12.227687",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.210546",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Remvove leakage in the test\n",
    "df_test = test_without_leakage"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "8994b339",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.247422Z",
     "iopub.status.busy": "2024-01-07T19:05:12.246775Z",
     "iopub.status.idle": "2024-01-07T19:05:12.280483Z",
     "shell.execute_reply": "2024-01-07T19:05:12.279703Z"
    },
    "papermill": {
     "duration": 0.046127,
     "end_time": "2024-01-07T19:05:12.282815",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.236688",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "DATA_DIR = \"/kaggle/input/data/\"\n",
    "NB_LABELS = len(unique_labels)\n",
    "BATCH_SIZE = 32\n",
    "\n",
    "X_train, X_valid = train_test_split(df_train, test_size = 0.2, random_state = 42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "16d2152b",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.302572Z",
     "iopub.status.busy": "2024-01-07T19:05:12.302235Z",
     "iopub.status.idle": "2024-01-07T19:05:12.316522Z",
     "shell.execute_reply": "2024-01-07T19:05:12.315636Z"
    },
    "papermill": {
     "duration": 0.026458,
     "end_time": "2024-01-07T19:05:12.318490",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.292032",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "class ChestImagesDataset(tf.keras.utils.Sequence):\n",
    "    def __init__(self, df, input_folder, batch_size=32, transform=None, training=True, shuffle=True):\n",
    "        self.df = df\n",
    "        self.input_folder = input_folder\n",
    "        self.batch_size = batch_size\n",
    "        self.transform = transform\n",
    "        self.training = training\n",
    "        self.shuffle = shuffle\n",
    "        \n",
    "    def on_epoch_end(self):\n",
    "        if self.shuffle:\n",
    "            self.df = self.df.sample(frac=1).reset_index(drop=True)\n",
    "    \n",
    "    def __len__(self):\n",
    "        return np.ceil(len(self.df) / self.batch_size).astype(int)\n",
    "    \n",
    "    def __getitem__(self, index):\n",
    "        batch = self.df[index * self.batch_size: (index + 1) * self.batch_size]\n",
    "        \n",
    "        images = []\n",
    "        \n",
    "        for filename in batch[\"Image Index\"]:\n",
    "            path = glob.glob(self.input_folder + \"*/*/\" + filename)[0]\n",
    "            image = Image.open(path).convert('RGB')\n",
    "            image = np.array(image)\n",
    "            \n",
    "            image = tf.image.resize(image, size=(224, 224))\n",
    "            image = tf.cast(image, dtype=tf.float32)\n",
    "            \n",
    "            images.append(image)\n",
    "        \n",
    "        images = np.array(images)\n",
    "        \n",
    "        # Normalize the images\n",
    "        #images = images / 255.0  # assuming the pixel values are in the range [0, 255]\n",
    "        \n",
    "        if self.training:\n",
    "            #labels = np.array(batch[unique_labels].values)\n",
    "            labels = np.array(batch[list(unique_labels)].values)\n",
    "            # Add mean to the images\n",
    "            images = images - np.mean(images, axis=(0, 1, 2), keepdims=True)\n",
    "            \n",
    "            return images, labels\n",
    "        else:\n",
    "            return images\n",
    "\n",
    "# Load the training data and the validation \n",
    "train_generator = ChestImagesDataset(X_train, DATA_DIR, batch_size=BATCH_SIZE, transform=None, shuffle=True)\n",
    "valid_generator = ChestImagesDataset(X_valid, DATA_DIR, batch_size=BATCH_SIZE, transform=None, shuffle=True)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "8556cd53",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.338075Z",
     "iopub.status.busy": "2024-01-07T19:05:12.337731Z",
     "iopub.status.idle": "2024-01-07T19:05:12.347728Z",
     "shell.execute_reply": "2024-01-07T19:05:12.346739Z"
    },
    "papermill": {
     "duration": 0.022022,
     "end_time": "2024-01-07T19:05:12.349638",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.327616",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "class MultiLabelCrossEntropy:\n",
    "    \"\"\"\n",
    "    Weight Cross Entropy Loss.\n",
    "    \n",
    "    For each class we will found two weights corresponding to the positive and negative frequency of our sample data.\n",
    "    These weights will manage the way we update our network. And this has for objectives to manage the unbalanced class issue.\n",
    "    \"\"\"\n",
    "        \n",
    "    def __init__(self, labels : DataFrame, epsilon = 1e-7):\n",
    "        \n",
    "        self.epsilon = epsilon\n",
    "        \n",
    "        # Get the size of the data\n",
    "        self.N = labels.shape[0]\n",
    "        \n",
    "        # Get the frequency occurence for each class\n",
    "        self.freq_pos = np.sum(labels == 1, axis=0) / self.N\n",
    "        self.freq_neg = np.sum(labels == 0, axis=0) / self.N\n",
    "        \n",
    "        # Set the loss weights for each labels \n",
    "        self.pos_weights = self.freq_neg\n",
    "        self.neg_weights = self.freq_pos\n",
    "        \n",
    "    def contribution(self):\n",
    "        \"\"\"\n",
    "        Get the weights' contribution for each labels.\n",
    "        \n",
    "        Returns :\n",
    "            - double : Positive contribution\n",
    "            - double : Negative contribution\n",
    "        \"\"\"\n",
    "        return self.freq_pos * self.pos_weights, self.freq_neg * self.neg_weights\n",
    "        \n",
    "        \n",
    "    def loss(self, y_true, y_pred):\n",
    "        \"\"\"\n",
    "        Return weighted loss value. \n",
    "        \"\"\"\n",
    "        # Initialize loss to zero\n",
    "        loss = 0.0\n",
    "        \n",
    "        for i in range(len(self.pos_weights)):\n",
    "            loss += (-1 * K.mean(\n",
    "                        self.pos_weights[i] * y_true[:,i] * K.log(y_pred[:,i] + self.epsilon))\n",
    "                    ) + (-1 * K.mean(\n",
    "                        self.neg_weights[i] * (1 - y_true[:,i]) * K.log(1 - y_pred[:,i] + self.epsilon))\n",
    "                    )\n",
    "        return loss\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "72290a8d",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.369502Z",
     "iopub.status.busy": "2024-01-07T19:05:12.368865Z",
     "iopub.status.idle": "2024-01-07T19:05:12.381298Z",
     "shell.execute_reply": "2024-01-07T19:05:12.380103Z"
    },
    "papermill": {
     "duration": 0.024756,
     "end_time": "2024-01-07T19:05:12.383606",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.358850",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "cross_entropy_loss = MultiLabelCrossEntropy(X_train[list(unique_labels)])\n",
    "pos_contribution, neg_contribution = cross_entropy_loss.contribution()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "8eedff59",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:12.404219Z",
     "iopub.status.busy": "2024-01-07T19:05:12.403519Z",
     "iopub.status.idle": "2024-01-07T19:05:25.993966Z",
     "shell.execute_reply": "2024-01-07T19:05:25.992886Z"
    },
    "papermill": {
     "duration": 13.603289,
     "end_time": "2024-01-07T19:05:25.996578",
     "exception": false,
     "start_time": "2024-01-07T19:05:12.393289",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting vit-keras\r\n",
      "  Downloading vit_keras-0.1.2-py3-none-any.whl (24 kB)\r\n",
      "Requirement already satisfied: scipy in /opt/conda/lib/python3.10/site-packages (from vit-keras) (1.11.4)\r\n",
      "Collecting validators (from vit-keras)\r\n",
      "  Obtaining dependency information for validators from https://files.pythonhosted.org/packages/3a/0c/785d317eea99c3739821718f118c70537639aa43f96bfa1d83a71f68eaf6/validators-0.22.0-py3-none-any.whl.metadata\r\n",
      "  Downloading validators-0.22.0-py3-none-any.whl.metadata (4.7 kB)\r\n",
      "Requirement already satisfied: numpy<1.28.0,>=1.21.6 in /opt/conda/lib/python3.10/site-packages (from scipy->vit-keras) (1.24.3)\r\n",
      "Downloading validators-0.22.0-py3-none-any.whl (26 kB)\r\n",
      "Installing collected packages: validators, vit-keras\r\n",
      "Successfully installed validators-0.22.0 vit-keras-0.1.2\r\n"
     ]
    }
   ],
   "source": [
    "!pip install vit-keras\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "6faa9324",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:26.017614Z",
     "iopub.status.busy": "2024-01-07T19:05:26.017234Z",
     "iopub.status.idle": "2024-01-07T19:05:32.894162Z",
     "shell.execute_reply": "2024-01-07T19:05:32.893312Z"
    },
    "papermill": {
     "duration": 6.890287,
     "end_time": "2024-01-07T19:05:32.896499",
     "exception": false,
     "start_time": "2024-01-07T19:05:26.006212",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://github.com/faustomorales/vit-keras/releases/download/dl/ViT-B_16_imagenet21k+imagenet2012.npz\n",
      "347502902/347502902 [==============================] - 1s 0us/step\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.10/site-packages/vit_keras/utils.py:81: UserWarning: Resizing position embeddings from 24, 24 to 14, 14\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "import tensorflow_addons as tfa\n",
    "from vit_keras import vit\n",
    "img_size=224\n",
    "vit_model = vit.vit_b16(\n",
    "        image_size = img_size,\n",
    "        activation = 'softmax',\n",
    "        pretrained = True,    \n",
    "        include_top = False,\n",
    "        pretrained_top = False,\n",
    "        classes = 14)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fb84debf",
   "metadata": {
    "papermill": {
     "duration": 0.012578,
     "end_time": "2024-01-07T19:05:32.921443",
     "exception": false,
     "start_time": "2024-01-07T19:05:32.908865",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "48061212",
   "metadata": {
    "papermill": {
     "duration": 0.011593,
     "end_time": "2024-01-07T19:05:32.944860",
     "exception": false,
     "start_time": "2024-01-07T19:05:32.933267",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c164b511",
   "metadata": {
    "papermill": {
     "duration": 0.011642,
     "end_time": "2024-01-07T19:05:32.968503",
     "exception": false,
     "start_time": "2024-01-07T19:05:32.956861",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e4226902",
   "metadata": {
    "papermill": {
     "duration": 0.011597,
     "end_time": "2024-01-07T19:05:32.991857",
     "exception": false,
     "start_time": "2024-01-07T19:05:32.980260",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c44a0f60",
   "metadata": {
    "papermill": {
     "duration": 0.011521,
     "end_time": "2024-01-07T19:05:33.015063",
     "exception": false,
     "start_time": "2024-01-07T19:05:33.003542",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "b071fd7e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:33.041588Z",
     "iopub.status.busy": "2024-01-07T19:05:33.040666Z",
     "iopub.status.idle": "2024-01-07T19:05:38.399206Z",
     "shell.execute_reply": "2024-01-07T19:05:38.398242Z"
    },
    "papermill": {
     "duration": 5.373689,
     "end_time": "2024-01-07T19:05:38.401424",
     "exception": false,
     "start_time": "2024-01-07T19:05:33.027735",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"vision_transformer\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " vit-b16 (Functional)        (None, 768)               85798656  \n",
      "                                                                 \n",
      " flatten (Flatten)           (None, 768)               0         \n",
      "                                                                 \n",
      " batch_normalization (Batch  (None, 768)               3072      \n",
      " Normalization)                                                  \n",
      "                                                                 \n",
      " dense (Dense)               (None, 256)               196864    \n",
      "                                                                 \n",
      " batch_normalization_1 (Bat  (None, 256)               1024      \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " dense_1 (Dense)             (None, 128)               32896     \n",
      "                                                                 \n",
      " batch_normalization_2 (Bat  (None, 128)               512       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " dense_2 (Dense)             (None, 64)                8256      \n",
      "                                                                 \n",
      " dense_3 (Dense)             (None, 15)                975       \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 86042255 (328.23 MB)\n",
      "Trainable params: 86039951 (328.22 MB)\n",
      "Non-trainable params: 2304 (9.00 KB)\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "def create_model():\n",
    "    vit_model = vit.vit_b16(\n",
    "        image_size=img_size,\n",
    "        activation='softmax',\n",
    "        pretrained=True,\n",
    "        include_top=False,\n",
    "        pretrained_top=False,\n",
    "        classes=14\n",
    "    )\n",
    "    model = tf.keras.Sequential([\n",
    "        vit_model,\n",
    "        tf.keras.layers.Flatten(),\n",
    "        tf.keras.layers.BatchNormalization(),\n",
    "        tf.keras.layers.Dense(256, activation=tfa.activations.gelu),\n",
    "        tf.keras.layers.BatchNormalization(),\n",
    "        tf.keras.layers.Dense(128, activation=tfa.activations.gelu),\n",
    "        tf.keras.layers.BatchNormalization(),\n",
    "        tf.keras.layers.Dense(64, activation=tfa.activations.gelu),\n",
    "        tf.keras.layers.Dense(15, activation='softmax')  # Adjust the number of units based on your specific needs\n",
    "    ],\n",
    "    name='vision_transformer')\n",
    "     \n",
    "    # Compile the model\n",
    "    model.compile(\n",
    "        optimizer=tf.keras.optimizers.Adam(),\n",
    "        loss='categorical_crossentropy', \n",
    "        metrics=['accuracy']\n",
    "    )\n",
    "    \n",
    "    return model, vit_model\n",
    "    \n",
    "\n",
    "# Create the model\n",
    "model, vit_model = create_model()\n",
    "\n",
    "# Print the summary\n",
    "model.summary()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "a2a12e79",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:05:38.431379Z",
     "iopub.status.busy": "2024-01-07T19:05:38.431047Z",
     "iopub.status.idle": "2024-01-07T19:19:28.907500Z",
     "shell.execute_reply": "2024-01-07T19:19:28.906551Z"
    },
    "papermill": {
     "duration": 830.494323,
     "end_time": "2024-01-07T19:19:28.909915",
     "exception": false,
     "start_time": "2024-01-07T19:05:38.415592",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4/4 [==============================] - 804s 251s/step - loss: 3.6710 - accuracy: 0.1875 - val_loss: 3.7301 - val_accuracy: 0.0156\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(train_generator, \n",
    "                    validation_data=valid_generator, \n",
    "                    epochs=1, \n",
    "                    steps_per_epoch=128 // BATCH_SIZE, \n",
    "                    batch_size=BATCH_SIZE)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "89e5a756",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T19:19:28.941600Z",
     "iopub.status.busy": "2024-01-07T19:19:28.940970Z",
     "iopub.status.idle": "2024-01-07T22:44:21.231169Z",
     "shell.execute_reply": "2024-01-07T22:44:21.228975Z"
    },
    "papermill": {
     "duration": 12292.837689,
     "end_time": "2024-01-07T22:44:21.762573",
     "exception": false,
     "start_time": "2024-01-07T19:19:28.924884",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/3\n",
      "1878/1878 [==============================] - 4431s 2s/step - loss: 25.8312 - accuracy: 0.3751 - val_loss: 83.9433 - val_accuracy: 0.4558\n",
      "Epoch 2/3\n",
      "1878/1878 [==============================] - 3821s 2s/step - loss: 139.0297 - accuracy: 0.2989 - val_loss: 365.7466 - val_accuracy: 0.0879\n",
      "Epoch 3/3\n",
      "1878/1878 [==============================] - 4039s 2s/step - loss: 322.2578 - accuracy: 0.2976 - val_loss: 335.9468 - val_accuracy: 0.2463\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(train_generator, validation_data=valid_generator, epochs = 3, batch_size = BATCH_SIZE) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "15d98ab4",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T22:44:22.702576Z",
     "iopub.status.busy": "2024-01-07T22:44:22.702141Z",
     "iopub.status.idle": "2024-01-07T22:44:58.729615Z",
     "shell.execute_reply": "2024-01-07T22:44:58.728808Z"
    },
    "papermill": {
     "duration": 36.499546,
     "end_time": "2024-01-07T22:44:58.732014",
     "exception": false,
     "start_time": "2024-01-07T22:44:22.232468",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Save the model \n",
    "model.save(\"model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "2f55ff0e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T22:44:59.678222Z",
     "iopub.status.busy": "2024-01-07T22:44:59.677841Z",
     "iopub.status.idle": "2024-01-07T22:45:00.045403Z",
     "shell.execute_reply": "2024-01-07T22:45:00.044564Z"
    },
    "papermill": {
     "duration": 0.843553,
     "end_time": "2024-01-07T22:45:00.047599",
     "exception": false,
     "start_time": "2024-01-07T22:44:59.204046",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "9d80f281",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T22:45:00.991263Z",
     "iopub.status.busy": "2024-01-07T22:45:00.990884Z",
     "iopub.status.idle": "2024-01-07T22:45:01.232707Z",
     "shell.execute_reply": "2024-01-07T22:45:01.231753Z"
    },
    "papermill": {
     "duration": 0.717505,
     "end_time": "2024-01-07T22:45:01.235005",
     "exception": false,
     "start_time": "2024-01-07T22:45:00.517500",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['accuracy'])\n",
    "plt.plot(history.history['val_accuracy'])\n",
    "\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "0c05f995",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2024-01-07T22:45:02.232870Z",
     "iopub.status.busy": "2024-01-07T22:45:02.231951Z",
     "iopub.status.idle": "2024-01-07T22:45:02.996056Z",
     "shell.execute_reply": "2024-01-07T22:45:02.994811Z"
    },
    "papermill": {
     "duration": 1.23159,
     "end_time": "2024-01-07T22:45:02.997631",
     "exception": true,
     "start_time": "2024-01-07T22:45:01.766041",
     "status": "failed"
    },
    "tags": []
   },
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "'auc'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[19], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m plt\u001b[38;5;241m.\u001b[39mplot(\u001b[43mhistory\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mhistory\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mauc\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m]\u001b[49m)\n\u001b[1;32m      2\u001b[0m plt\u001b[38;5;241m.\u001b[39mplot(history\u001b[38;5;241m.\u001b[39mhistory[\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mval_auc\u001b[39m\u001b[38;5;124m'\u001b[39m])\n\u001b[1;32m      4\u001b[0m plt\u001b[38;5;241m.\u001b[39mtitle(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mmodel auc\u001b[39m\u001b[38;5;124m'\u001b[39m)\n",
      "\u001b[0;31mKeyError\u001b[0m: 'auc'"
     ]
    }
   ],
   "source": [
    "plt.plot(history.history['auc'])\n",
    "plt.plot(history.history['val_auc'])\n",
    "\n",
    "plt.title('model auc')\n",
    "plt.ylabel('auc')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "32245c49",
   "metadata": {
    "execution": {
     "iopub.status.busy": "2024-01-07T18:59:07.807498Z",
     "iopub.status.idle": "2024-01-07T18:59:07.807857Z",
     "shell.execute_reply": "2024-01-07T18:59:07.807701Z",
     "shell.execute_reply.started": "2024-01-07T18:59:07.807684Z"
    },
    "papermill": {
     "duration": null,
     "end_time": null,
     "exception": null,
     "start_time": null,
     "status": "pending"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "test_generator = ChestImagesDataset(df_test, DATA_DIR, batch_size = BATCH_SIZE, transform = None, training = False, shuffle = False)\n",
    "\n",
    "predictions = model.predict(test_generator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f8a277e8",
   "metadata": {
    "execution": {
     "iopub.status.busy": "2024-01-07T18:59:07.809668Z",
     "iopub.status.idle": "2024-01-07T18:59:07.810128Z",
     "shell.execute_reply": "2024-01-07T18:59:07.809914Z",
     "shell.execute_reply.started": "2024-01-07T18:59:07.809893Z"
    },
    "papermill": {
     "duration": null,
     "end_time": null,
     "exception": null,
     "start_time": null,
     "status": "pending"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "roc_auc_score(df_test[unique_labels].values, predictions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b02eaf93",
   "metadata": {
    "execution": {
     "iopub.status.busy": "2024-01-07T18:59:07.811837Z",
     "iopub.status.idle": "2024-01-07T18:59:07.812597Z",
     "shell.execute_reply": "2024-01-07T18:59:07.812338Z",
     "shell.execute_reply.started": "2024-01-07T18:59:07.812315Z"
    },
    "papermill": {
     "duration": null,
     "end_time": null,
     "exception": null,
     "start_time": null,
     "status": "pending"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "## Metrics\n",
    "\n",
    "def get_roc_curve(labels, predicted_vals, real_vals):\n",
    "    auc_roc_vals = []\n",
    "    for i in range(len(labels)):\n",
    "        try:\n",
    "            gt = real_vals[:, i]\n",
    "            pred = predicted_vals[:, i]\n",
    "            auc_roc = roc_auc_score(gt, pred)\n",
    "            auc_roc_vals.append(auc_roc)\n",
    "            fpr_rf, tpr_rf, _ = roc_curve(gt, pred)\n",
    "            plt.figure(1, figsize=(10, 10))\n",
    "            plt.plot([0, 1], [0, 1], 'k--')\n",
    "            plt.plot(fpr_rf, tpr_rf,\n",
    "                     label=labels[i] + \" (\" + str(round(auc_roc, 3)) + \")\")\n",
    "            plt.xlabel('False positive rate')\n",
    "            plt.ylabel('True positive rate')\n",
    "            plt.title('ROC curve')\n",
    "            plt.legend(loc='best')\n",
    "        except:\n",
    "            print(\n",
    "                f\"Error in generating ROC curve for {labels[i]}. \"\n",
    "                f\"Dataset lacks enough examples.\"\n",
    "            )\n",
    "    plt.show()\n",
    "    return auc_roc_vals\n",
    "\n",
    "\n",
    "get_roc_curve(list(unique_labels), predictions, df_test[unique_labels].values)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "717c487e",
   "metadata": {
    "execution": {
     "iopub.status.busy": "2024-01-07T18:59:07.814058Z",
     "iopub.status.idle": "2024-01-07T18:59:07.814555Z",
     "shell.execute_reply": "2024-01-07T18:59:07.814317Z",
     "shell.execute_reply.started": "2024-01-07T18:59:07.814294Z"
    },
    "papermill": {
     "duration": null,
     "end_time": null,
     "exception": null,
     "start_time": null,
     "status": "pending"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Grad-CAM\n",
    "class GradCAM:\n",
    "  def __init__(self, model, classIdx, baseModel = None, layerName = None):\n",
    "    # store the model, the class index used to measure the class\n",
    "    # activation map, and the layer to be used when visualizing\n",
    "    # the class activation map\n",
    "    if baseModel:\n",
    "      self.model = model.get_layer(baseModel)\n",
    "    else:\n",
    "      self.model = model\n",
    "    self.classIdx = classIdx\n",
    "    self.layerName = layerName\n",
    "    # if the layer name is None, attempt to automatically find\n",
    "    # the target output layer\n",
    "    if self.layerName is None:\n",
    "      self.layerName = self.find_target_layer()\n",
    "\n",
    "  def find_target_layer(self):\n",
    "    # attempt to find the final convolutional layer in the network\n",
    "    # by looping over the layers of the network in reverse order\n",
    "    for layer in reversed(self.model.layers):\n",
    "        # check to see if the layer has a 4D output\n",
    "        if len(layer.output_shape) == 4:\n",
    "            return layer.name\n",
    "    # otherwise, we could not find a 4D layer so the GradCAM\n",
    "    # algorithm cannot be applied\n",
    "    raise ValueError(\"Could not find 4D layer. Cannot apply GradCAM.\")\n",
    "\n",
    "  def compute_heatmap(self, image, eps=1e-8):\n",
    "    # construct our gradient model by supplying (1) the inputs\n",
    "    # to our pre-trained model, (2) the output of the (presumably)\n",
    "    # final 4D layer in the network, and (3) the output of the\n",
    "    # softmax activations from the model\n",
    "    gradModel = keras.Model(\n",
    "      inputs=[self.model.inputs],\n",
    "      outputs=[self.model.get_layer(self.layerName).output,\n",
    "        self.model.output])\n",
    "\n",
    "    # record operations for automatic differentiation\n",
    "    with tf.GradientTape() as tape:\n",
    "      # cast the image tensor to a float-32 data type, pass the\n",
    "      # image through the gradient model, and grab the loss\n",
    "      # associated with the specific class index\n",
    "      inputs = tf.cast(image, tf.float32)\n",
    "      (convOutputs, predictions) = gradModel(inputs)\n",
    "      loss = predictions[:, self.classIdx]\n",
    "    # use automatic differentiation to compute the gradients\n",
    "    grads = tape.gradient(loss, convOutputs)\n",
    "\n",
    "    # compute the guided gradients\n",
    "    castConvOutputs = tf.cast(convOutputs > 0, \"float32\")\n",
    "    castGrads = tf.cast(grads > 0, \"float32\")\n",
    "    guidedGrads = castConvOutputs * castGrads * grads\n",
    "    # the convolution and guided gradients have a batch dimension\n",
    "    # (which we don't need) so let's grab the volume itself and\n",
    "    # discard the batch\n",
    "    convOutputs = convOutputs[0]\n",
    "    guidedGrads = guidedGrads[0]\n",
    "\n",
    "    # compute the average of the gradient values, and using them\n",
    "    # as weights, compute the ponderation of the filters with\n",
    "    # respect to the weights\n",
    "    weights = tf.reduce_mean(guidedGrads, axis=(0, 1))\n",
    "    cam = tf.reduce_sum(tf.multiply(weights, convOutputs), axis=-1)\n",
    "\n",
    "    # grab the spatial dimensions of the input image and resize\n",
    "    # the output class activation map to match the input image\n",
    "    # dimensions\n",
    "    (w, h) = (image.shape[2], image.shape[1])\n",
    "    heatmap = cv2.resize(cam.numpy(), (w, h))\n",
    "    # normalize the heatmap such that all values lie in the range\n",
    "    # [0, 1], scale the resulting values to the range [0, 255],\n",
    "    # and then convert to an unsigned 8-bit integer\n",
    "    numer = heatmap - np.min(heatmap)\n",
    "    denom = (heatmap.max() - heatmap.min()) + eps\n",
    "    heatmap = numer / denom\n",
    "    heatmap = (heatmap * 255).astype(\"uint8\")\n",
    "    # return the resulting heatmap to the calling function\n",
    "    return heatmap\n",
    "\n",
    "  def overlay_heatmap(self, heatmap, image, alpha=0.5, colormap=cv2.COLORMAP_VIRIDIS):\n",
    "    # apply the supplied color map to the heatmap and then\n",
    "    # overlay the heatmap on the input image\n",
    "    image = img_as_ubyte(image)\n",
    "    heatmap = cv2.applyColorMap(heatmap, colormap)\n",
    "    output = cv2.addWeighted(image, alpha, heatmap, 1 - alpha, 0)\n",
    "    # return a 2-tuple of the color mapped heatmap and the output,\n",
    "    # overlaid image\n",
    "    return (heatmap, output)\n",
    "\n",
    "\n",
    "idx = 45\n",
    "image = X_test[idx:(idx+1),:,:,:]\n",
    "preds = model.predict(image)\n",
    "i = np.argmax(preds[0])\n",
    "\n",
    "cam = GradCAM(model, i, baseModel = \"vit-b16\")#, layerName = \"mixed10\")\n",
    "heatmap = cam.compute_heatmap(image)\n",
    "heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_VIRIDIS)\n",
    "(heatmap, output) = cam.overlay_heatmap(heatmap, image[0,:,:,:], alpha = 0.3)\n",
    "plt.imshow(output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8db5c081",
   "metadata": {
    "execution": {
     "iopub.status.busy": "2024-01-07T18:59:07.815543Z",
     "iopub.status.idle": "2024-01-07T18:59:07.816005Z",
     "shell.execute_reply": "2024-01-07T18:59:07.815794Z",
     "shell.execute_reply.started": "2024-01-07T18:59:07.815773Z"
    },
    "papermill": {
     "duration": null,
     "end_time": null,
     "exception": null,
     "start_time": null,
     "status": "pending"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "gradcam_explain = np.zeros((3, 224, 224))\n",
    "\n",
    "nimages, _, _, _ = X_test.shape\n",
    "with tf.device(\"/gpu:1\"):\n",
    "  for n in range(nimages):\n",
    "    image = X_test[n,:,:,:].reshape(1, 224, 224, 3)\n",
    "    preds = model.predict(image)\n",
    "    top_label = np.argmax(preds[0])\n",
    "\n",
    "    cam = GradCAM(model, top_label, baseModel = \"vit-b16\")#, layerName = \"mixed10\")\n",
    "    heatmap = cam.compute_heatmap(image)\n",
    "    heatmap = heatmap / 255\n",
    "    \n",
    "    gradcam_explain[top_label,:,:] = np.add(gradcam_explain[top_label,:,:], heatmap)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "db85e2ce",
   "metadata": {
    "execution": {
     "iopub.status.busy": "2024-01-07T18:59:07.817640Z",
     "iopub.status.idle": "2024-01-07T18:59:07.817983Z",
     "shell.execute_reply": "2024-01-07T18:59:07.817835Z",
     "shell.execute_reply.started": "2024-01-07T18:59:07.817819Z"
    },
    "papermill": {
     "duration": null,
     "end_time": null,
     "exception": null,
     "start_time": null,
     "status": "pending"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(20, 20)\n",
    "sns.heatmap(gradcam_explain[0,:,:], cmap = \"YlOrRd\", xticklabels = False, yticklabels = False, cbar = False)"
   ]
  }
 ],
 "metadata": {
  "kaggle": {
   "accelerator": "gpu",
   "dataSources": [
    {
     "datasetId": 5839,
     "sourceId": 18613,
     "sourceType": "datasetVersion"
    }
   ],
   "dockerImageVersionId": 30627,
   "isGpuEnabled": true,
   "isInternetEnabled": true,
   "language": "python",
   "sourceType": "notebook"
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  },
  "papermill": {
   "default_parameters": {},
   "duration": 13226.208746,
   "end_time": "2024-01-07T22:45:06.730318",
   "environment_variables": {},
   "exception": true,
   "input_path": "__notebook__.ipynb",
   "output_path": "__notebook__.ipynb",
   "parameters": {},
   "start_time": "2024-01-07T19:04:40.521572",
   "version": "2.4.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}