evaluate_finetune.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import torch\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import StratifiedGroupKFold\n",
    "from torch_dataset import BreastCancerDataSet_Basic\n",
    "from transforms import get_transforms\n",
    "from training import train_model\n",
    "from models import SimSiam, SimSiamWithAffine\n",
    "import gc\n",
    "import numpy as np\n",
    "from training import extract_model_features\n",
    "\n",
    "N_FOLDS = 5\n",
    "IMG_SIZE = (256, 512)\n",
    "MODEL_TYPE = \"resnet50\"\n",
    "TARGET = 'cancer'\n",
    "DEVICE = \"cuda:0\"\n",
    "\n",
    "train_path = \"/mnt/striped/kaggle/rsna_bcd/data/train.csv\"\n",
    "images_path = \"/mnt/striped/kaggle/rsna_bcd/data/preprocessed_mp_aspectratio_768_384/\"\n",
    "\n",
    "os.environ[\"NCCL_BUFFSIZE\"] = f\"{64 * 1024 * 1024}\"\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>site_id</th>\n",
       "      <th>patient_id</th>\n",
       "      <th>image_id</th>\n",
       "      <th>laterality</th>\n",
       "      <th>view</th>\n",
       "      <th>age</th>\n",
       "      <th>cancer</th>\n",
       "      <th>biopsy</th>\n",
       "      <th>invasive</th>\n",
       "      <th>BIRADS</th>\n",
       "      <th>implant</th>\n",
       "      <th>density</th>\n",
       "      <th>machine_id</th>\n",
       "      <th>difficult_negative_case</th>\n",
       "      <th>split</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2</td>\n",
       "      <td>10006</td>\n",
       "      <td>462822612</td>\n",
       "      <td>L</td>\n",
       "      <td>CC</td>\n",
       "      <td>61.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>29</td>\n",
       "      <td>False</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>10006</td>\n",
       "      <td>1459541791</td>\n",
       "      <td>L</td>\n",
       "      <td>MLO</td>\n",
       "      <td>61.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>29</td>\n",
       "      <td>False</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>10006</td>\n",
       "      <td>1864590858</td>\n",
       "      <td>R</td>\n",
       "      <td>MLO</td>\n",
       "      <td>61.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>29</td>\n",
       "      <td>False</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2</td>\n",
       "      <td>10006</td>\n",
       "      <td>1874946579</td>\n",
       "      <td>R</td>\n",
       "      <td>CC</td>\n",
       "      <td>61.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>29</td>\n",
       "      <td>False</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2</td>\n",
       "      <td>10011</td>\n",
       "      <td>220375232</td>\n",
       "      <td>L</td>\n",
       "      <td>CC</td>\n",
       "      <td>55.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>21</td>\n",
       "      <td>True</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   site_id  patient_id    image_id laterality view   age  cancer  biopsy  \\\n",
       "0        2       10006   462822612          L   CC  61.0       0       0   \n",
       "1        2       10006  1459541791          L  MLO  61.0       0       0   \n",
       "2        2       10006  1864590858          R  MLO  61.0       0       0   \n",
       "3        2       10006  1874946579          R   CC  61.0       0       0   \n",
       "4        2       10011   220375232          L   CC  55.0       0       0   \n",
       "\n",
       "   invasive  BIRADS  implant density  machine_id  difficult_negative_case  \\\n",
       "0         0     NaN        0     NaN          29                    False   \n",
       "1         0     NaN        0     NaN          29                    False   \n",
       "2         0     NaN        0     NaN          29                    False   \n",
       "3         0     NaN        0     NaN          29                    False   \n",
       "4         0     0.0        0     NaN          21                     True   \n",
       "\n",
       "   split  \n",
       "0      1  \n",
       "1      1  \n",
       "2      1  \n",
       "3      1  \n",
       "4      0  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Load dataset\n",
    "df_train = pd.read_csv(train_path)\n",
    "\n",
    "# Set up the folds\n",
    "split = StratifiedGroupKFold(N_FOLDS, random_state=42, shuffle=True)\n",
    "for k, (_, test_idx) in enumerate(split.split(df_train, df_train.cancer, groups=df_train.patient_id)):\n",
    "    df_train.loc[test_idx, 'split'] = k\n",
    "df_train.split = df_train.split.astype(int)\n",
    "df_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_model(model_type, affine, mixup, checkpoint, device):\n",
    "    \n",
    "    if affine:\n",
    "        if mixup:\n",
    "            raise ValueError(\"Affine and mixup are not compatible\")\n",
    "\n",
    "        model = SimSiamWithAffine(model_type = model_type, IMG_SIZE = IMG_SIZE)\n",
    "    else:\n",
    "        model = SimSiam(model_type = model_type, IMG_SIZE = IMG_SIZE)\n",
    "\n",
    "    model.to(device)\n",
    "    model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)\n",
    "\n",
    "    def adjust_key(key : str):\n",
    "        if key.startswith(\"module.\"):\n",
    "            return key[7:]\n",
    "        else:\n",
    "            return key\n",
    "        \n",
    "    state_dict = torch.load(checkpoint)\n",
    "    state_dict = {adjust_key(k):v for k,v in state_dict.items()}\n",
    "    model.load_state_dict(state_dict)\n",
    "\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "from torch.cuda.amp import GradScaler, autocast\n",
    "from sklearn.metrics import roc_auc_score\n",
    "\n",
    "def pfbeta(labels, predictions, beta):\n",
    "    y_true_count = 0\n",
    "    ctp = 0\n",
    "    cfp = 0\n",
    "\n",
    "    for idx in range(len(labels)):\n",
    "        prediction = min(max(predictions[idx], 0), 1)\n",
    "        if (labels[idx]):\n",
    "            y_true_count += 1\n",
    "            ctp += prediction\n",
    "        else:\n",
    "            cfp += prediction\n",
    "\n",
    "    beta_squared = beta * beta\n",
    "    c_precision = ctp / (ctp + cfp)\n",
    "    c_recall = ctp / y_true_count\n",
    "    if (c_precision > 0 and c_recall > 0):\n",
    "        result = (1 + beta_squared) * (c_precision * c_recall) / (beta_squared * c_precision + c_recall)\n",
    "        return result\n",
    "    else:\n",
    "        return 0\n",
    "    \n",
    "def train_model_simple(model, ds_train, ds_eval,\n",
    "                batch_size, device, epochs, sampler = None, \n",
    "                shuffle = True):\n",
    "    \n",
    "    num_workers = 6\n",
    "    \n",
    "    torch.manual_seed(42)\n",
    "        \n",
    "    print(\"Device\", device)\n",
    "    print(\"Setting up loader ...\")\n",
    "\n",
    "    dl_train = torch.utils.data.DataLoader(ds_train, batch_size=batch_size, shuffle=shuffle, num_workers=num_workers, \n",
    "                                           pin_memory=True, pin_memory_device=device, prefetch_factor=4, sampler=sampler)\n",
    "\n",
    "    dl_eval = torch.utils.data.DataLoader(ds_eval, batch_size=batch_size, shuffle=False, num_workers=num_workers, \n",
    "                                           pin_memory=True, pin_memory_device=device, prefetch_factor=4, sampler=None)\n",
    "\n",
    "    print(\"Setting up loader ... DONE\")\n",
    "\n",
    "    criterion = nn.BCEWithLogitsLoss()\n",
    "\n",
    "    # use SGD with momentum and weight decay\n",
    "    optimizer = torch.optim.SGD(model.parameters(), lr=0.001, momentum=0.9, weight_decay=5e-4)\n",
    "    scheduler = None\n",
    "    \n",
    "    scaler = GradScaler()\n",
    "    best_eval_score = 0\n",
    "\n",
    "    avg_loss = 0.0\n",
    "    aucs = []\n",
    "\n",
    "    for epoch in range(epochs):\n",
    "\n",
    "        model.train()\n",
    "            \n",
    "        optimizer.zero_grad(set_to_none=True)\n",
    "        \n",
    "        for x, y_cancer in dl_train:\n",
    "\n",
    "            # Using mixed precision training\n",
    "            with autocast():\n",
    "                x = x.to(device)\n",
    "                y_cancer = y_cancer.to(device)\n",
    "                y_pred = model(x)\n",
    "                loss = criterion(y_pred, y_cancer.float())\n",
    "                \n",
    "            # scaler is needed to prevent \"gradient underflow\"\n",
    "            scaler.scale(loss).backward()\n",
    "            scaler.step(optimizer)\n",
    "            scaler.update()\n",
    "            optimizer.zero_grad()\n",
    "            \n",
    "            # use moving averages to track the loss\n",
    "            w = 0.9\n",
    "            avg_loss = w * avg_loss + (1 - w) * loss.item()\n",
    "             \n",
    "        # print intermediate results\n",
    "        print(\n",
    "            f\"[Epoch {epoch:3d}] \"\n",
    "            f\"Loss = {avg_loss:.4f} \"\n",
    "        )\n",
    "        \n",
    "        if epoch % 2 == 0:\n",
    "        \n",
    "            model.eval()\n",
    "\n",
    "            eval_preds = []\n",
    "            eval_targets = []\n",
    "\n",
    "            with torch.no_grad():\n",
    "\n",
    "                for x, y_cancer in dl_eval:\n",
    "\n",
    "                    x = x.to(device)\n",
    "                    y_pred = model(x).squeeze()\n",
    "                    y_pred = torch.sigmoid(y_pred)\n",
    "\n",
    "                    eval_preds.extend(y_pred.tolist())\n",
    "                    eval_targets.extend(y_cancer.tolist())\n",
    "\n",
    "            auc = roc_auc_score(eval_targets, eval_preds)\n",
    "            pf1= pfbeta(eval_targets, eval_preds, beta=1)\n",
    "                        \n",
    "            aucs.append(auc)\n",
    "\n",
    "            print(f'AUC on test set: {auc:.2f}')\n",
    "            print(f'PF1 on test set: {pf1:.2f}')\n",
    "\n",
    "    return aucs\n",
    "            \n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Device cuda:0\n",
      "Setting up loader ...\n",
      "Setting up loader ... DONE\n",
      "[Epoch   0] Loss = 0.0994 \n",
      "AUC on test set: 0.55\n",
      "PF1 on test set: 0.02\n",
      "[Epoch   1] Loss = 0.0828 \n",
      "[Epoch   2] Loss = 0.0650 \n",
      "AUC on test set: 0.58\n",
      "PF1 on test set: 0.02\n",
      "[Epoch   3] Loss = 0.0933 \n",
      "[Epoch   4] Loss = 0.0945 \n",
      "AUC on test set: 0.61\n",
      "PF1 on test set: 0.03\n",
      "[Epoch   5] Loss = 0.0934 \n",
      "[Epoch   6] Loss = 0.0967 \n",
      "AUC on test set: 0.60\n",
      "PF1 on test set: 0.03\n",
      "[Epoch   7] Loss = 0.0871 \n",
      "[Epoch   8] Loss = 0.0847 \n",
      "AUC on test set: 0.62\n",
      "PF1 on test set: 0.03\n",
      "[Epoch   9] Loss = 0.0692 \n",
      "[Epoch  10] Loss = 0.1027 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  11] Loss = 0.0812 \n",
      "[Epoch  12] Loss = 0.0769 \n",
      "AUC on test set: 0.65\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  13] Loss = 0.0955 \n",
      "[Epoch  14] Loss = 0.0839 \n",
      "AUC on test set: 0.63\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  15] Loss = 0.1049 \n",
      "[Epoch  16] Loss = 0.0956 \n",
      "AUC on test set: 0.67\n",
      "PF1 on test set: 0.04\n",
      "[Epoch  17] Loss = 0.0922 \n",
      "[Epoch  18] Loss = 0.1385 \n",
      "AUC on test set: 0.67\n",
      "PF1 on test set: 0.04\n",
      "[Epoch  19] Loss = 0.0928 \n",
      "[Epoch  20] Loss = 0.0988 \n",
      "AUC on test set: 0.63\n",
      "PF1 on test set: 0.04\n",
      "[Epoch  21] Loss = 0.1118 \n",
      "[Epoch  22] Loss = 0.1143 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  23] Loss = 0.0873 \n",
      "[Epoch  24] Loss = 0.0954 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  25] Loss = 0.0621 \n",
      "[Epoch  26] Loss = 0.0749 \n",
      "AUC on test set: 0.63\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  27] Loss = 0.1036 \n",
      "[Epoch  28] Loss = 0.0798 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  29] Loss = 0.0859 \n",
      "[Epoch  30] Loss = 0.0862 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  31] Loss = 0.0797 \n",
      "[Epoch  32] Loss = 0.1058 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  33] Loss = 0.1130 \n",
      "[Epoch  34] Loss = 0.1483 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  35] Loss = 0.1201 \n",
      "[Epoch  36] Loss = 0.1136 \n",
      "AUC on test set: 0.65\n",
      "PF1 on test set: 0.04\n",
      "[Epoch  37] Loss = 0.1037 \n",
      "[Epoch  38] Loss = 0.1272 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.03\n",
      "[Epoch  39] Loss = 0.0874 \n",
      "Device cuda:0\n",
      "Setting up loader ...\n",
      "Setting up loader ... DONE\n",
      "[Epoch   0] Loss = 0.1044 \n",
      "AUC on test set: 0.57\n",
      "PF1 on test set: 0.02\n",
      "[Epoch   1] Loss = 0.0814 \n",
      "[Epoch   2] Loss = 0.0662 \n",
      "AUC on test set: 0.62\n",
      "PF1 on test set: 0.03\n",
      "[Epoch   3] Loss = 0.0886 \n",
      "[Epoch   4] Loss = 0.0904 \n",
      "AUC on test set: 0.63\n",
      "PF1 on test set: 0.03\n",
      "[Epoch   5] Loss = 0.0879 \n",
      "[Epoch   6] Loss = 0.0897 \n",
      "AUC on test set: 0.67\n",
      "PF1 on test set: 0.06\n",
      "[Epoch   7] Loss = 0.0785 \n",
      "[Epoch   8] Loss = 0.0826 \n",
      "AUC on test set: 0.68\n",
      "PF1 on test set: 0.07\n",
      "[Epoch   9] Loss = 0.0559 \n",
      "[Epoch  10] Loss = 0.0843 \n",
      "AUC on test set: 0.70\n",
      "PF1 on test set: 0.08\n",
      "[Epoch  11] Loss = 0.0668 \n",
      "[Epoch  12] Loss = 0.0613 \n",
      "AUC on test set: 0.70\n",
      "PF1 on test set: 0.09\n",
      "[Epoch  13] Loss = 0.0920 \n",
      "[Epoch  14] Loss = 0.0708 \n",
      "AUC on test set: 0.71\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  15] Loss = 0.0812 \n",
      "[Epoch  16] Loss = 0.0661 \n",
      "AUC on test set: 0.71\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  17] Loss = 0.0644 \n",
      "[Epoch  18] Loss = 0.1112 \n",
      "AUC on test set: 0.69\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  19] Loss = 0.0712 \n",
      "[Epoch  20] Loss = 0.0676 \n",
      "AUC on test set: 0.67\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  21] Loss = 0.0839 \n",
      "[Epoch  22] Loss = 0.0818 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.08\n",
      "[Epoch  23] Loss = 0.0629 \n",
      "[Epoch  24] Loss = 0.0613 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  25] Loss = 0.0417 \n",
      "[Epoch  26] Loss = 0.0476 \n",
      "AUC on test set: 0.63\n",
      "PF1 on test set: 0.09\n",
      "[Epoch  27] Loss = 0.0574 \n",
      "[Epoch  28] Loss = 0.0428 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.09\n",
      "[Epoch  29] Loss = 0.0427 \n",
      "[Epoch  30] Loss = 0.0483 \n",
      "AUC on test set: 0.67\n",
      "PF1 on test set: 0.09\n",
      "[Epoch  31] Loss = 0.0448 \n",
      "[Epoch  32] Loss = 0.0410 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  33] Loss = 0.0447 \n",
      "[Epoch  34] Loss = 0.0802 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  35] Loss = 0.0646 \n",
      "[Epoch  36] Loss = 0.0568 \n",
      "AUC on test set: 0.63\n",
      "PF1 on test set: 0.08\n",
      "[Epoch  37] Loss = 0.0481 \n",
      "[Epoch  38] Loss = 0.0482 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.12\n",
      "[Epoch  39] Loss = 0.0366 \n",
      "Device cuda:0\n",
      "Setting up loader ...\n",
      "Setting up loader ... DONE\n",
      "[Epoch   0] Loss = 0.1014 \n",
      "AUC on test set: 0.61\n",
      "PF1 on test set: 0.02\n",
      "[Epoch   1] Loss = 0.0793 \n",
      "[Epoch   2] Loss = 0.0639 \n",
      "AUC on test set: 0.67\n",
      "PF1 on test set: 0.03\n",
      "[Epoch   3] Loss = 0.0883 \n",
      "[Epoch   4] Loss = 0.0888 \n",
      "AUC on test set: 0.69\n",
      "PF1 on test set: 0.05\n",
      "[Epoch   5] Loss = 0.0859 \n",
      "[Epoch   6] Loss = 0.0883 \n",
      "AUC on test set: 0.71\n",
      "PF1 on test set: 0.06\n",
      "[Epoch   7] Loss = 0.0779 \n",
      "[Epoch   8] Loss = 0.0756 \n",
      "AUC on test set: 0.74\n",
      "PF1 on test set: 0.09\n",
      "[Epoch   9] Loss = 0.0546 \n",
      "[Epoch  10] Loss = 0.0807 \n",
      "AUC on test set: 0.73\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  11] Loss = 0.0726 \n",
      "[Epoch  12] Loss = 0.0627 \n",
      "AUC on test set: 0.70\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  13] Loss = 0.0854 \n",
      "[Epoch  14] Loss = 0.0665 \n",
      "AUC on test set: 0.74\n",
      "PF1 on test set: 0.11\n",
      "[Epoch  15] Loss = 0.0809 \n",
      "[Epoch  16] Loss = 0.0622 \n",
      "AUC on test set: 0.73\n",
      "PF1 on test set: 0.12\n",
      "[Epoch  17] Loss = 0.0591 \n",
      "[Epoch  18] Loss = 0.0985 \n",
      "AUC on test set: 0.73\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  19] Loss = 0.0576 \n",
      "[Epoch  20] Loss = 0.0715 \n",
      "AUC on test set: 0.68\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  21] Loss = 0.0648 \n",
      "[Epoch  22] Loss = 0.0719 \n",
      "AUC on test set: 0.69\n",
      "PF1 on test set: 0.09\n",
      "[Epoch  23] Loss = 0.0586 \n",
      "[Epoch  24] Loss = 0.0565 \n",
      "AUC on test set: 0.69\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  25] Loss = 0.0409 \n",
      "[Epoch  26] Loss = 0.0419 \n",
      "AUC on test set: 0.66\n",
      "PF1 on test set: 0.08\n",
      "[Epoch  27] Loss = 0.0477 \n",
      "[Epoch  28] Loss = 0.0396 \n",
      "AUC on test set: 0.68\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  29] Loss = 0.0454 \n",
      "[Epoch  30] Loss = 0.0404 \n",
      "AUC on test set: 0.69\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  31] Loss = 0.0514 \n",
      "[Epoch  32] Loss = 0.0531 \n",
      "AUC on test set: 0.68\n",
      "PF1 on test set: 0.10\n",
      "[Epoch  33] Loss = 0.0516 \n",
      "[Epoch  34] Loss = 0.0616 \n",
      "AUC on test set: 0.62\n",
      "PF1 on test set: 0.07\n",
      "[Epoch  35] Loss = 0.0460 \n",
      "[Epoch  36] Loss = 0.0483 \n",
      "AUC on test set: 0.64\n",
      "PF1 on test set: 0.05\n",
      "[Epoch  37] Loss = 0.0426 \n",
      "[Epoch  38] Loss = 0.0548 \n",
      "AUC on test set: 0.69\n",
      "PF1 on test set: 0.13\n",
      "[Epoch  39] Loss = 0.0311 \n"
     ]
    }
   ],
   "source": [
    "import torch.nn as nn\n",
    "from timm import create_model\n",
    "\n",
    "def finetune_model(model_backbone, backbone_dim, epochs, device):\n",
    "\n",
    "    # Create a simple binary classifier\n",
    "    class BinaryClassificationModel(nn.Module):\n",
    "\n",
    "        def __init__(self, backbone, backbone_dim):\n",
    "            super(BinaryClassificationModel, self).__init__()\n",
    "            self.backbone = backbone\n",
    "\n",
    "            # Classification head\n",
    "            self.classifier = nn.Sequential(\n",
    "                nn.BatchNorm1d(backbone_dim),\n",
    "                nn.ReLU(),\n",
    "                nn.Linear(backbone_dim, 1)\n",
    "            )\n",
    "        \n",
    "        def forward(self, x):\n",
    "            # Get the features from the backbone\n",
    "            features = self.backbone(x)\n",
    "            # Classify\n",
    "            output = self.classifier(features)\n",
    "            output = output.squeeze()\n",
    "\n",
    "            return output\n",
    "        \n",
    "    model_classifier = BinaryClassificationModel(model_backbone, backbone_dim)\n",
    "    model_classifier = model_classifier.to(device)\n",
    "\n",
    "    fold = 0\n",
    "\n",
    "    ds_train = BreastCancerDataSet_Basic(df_train.query('split != @fold'), \n",
    "                                        images_path,\n",
    "                                        TARGET = TARGET,\n",
    "                                        transforms = get_transforms(True, IMG_SIZE))\n",
    "\n",
    "    ds_eval = BreastCancerDataSet_Basic(df_train.query('split == @fold'), \n",
    "                                        images_path,\n",
    "                                        TARGET = TARGET,\n",
    "                                        transforms = get_transforms(False, IMG_SIZE))\n",
    "\n",
    "    return train_model_simple(model_classifier, ds_train, ds_eval, 64, device, epochs)\n",
    "\n",
    "epochs = 40\n",
    "\n",
    "# Perform three different fine-tuning experiments: imagenet weights, normal weights, mixup weights\n",
    "auc_finetune_none   = finetune_model(create_model(MODEL_TYPE, pretrained = True, num_classes = 0, drop_rate = 0., in_chans = 1), 2048, epochs, DEVICE)\n",
    "auc_finetune_normal = finetune_model(load_model(MODEL_TYPE, False, False, \"checkpoints/model_resnet50_bs_48_epoch_90.pt\", DEVICE).backbone, 2048, epochs, DEVICE)\n",
    "auc_finetune_mixup  = finetune_model(load_model(MODEL_TYPE, False, True,\"checkpoints/model_resnet50_mixup_bs_48_epoch_90.pt\", DEVICE).backbone, 2048, epochs, DEVICE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "l = 10\n",
    "x = range(l)\n",
    "\n",
    "plt.plot(x, auc_finetune_none[:l], label='None')\n",
    "#plt.plot(x, auc_finetune_affine, label='Affine')\n",
    "plt.plot(x, auc_finetune_normal[:l], label='Normal')\n",
    "plt.plot(x, auc_finetune_mixup[:l], label='Mixup')\n",
    "\n",
    "# Adding title and labels\n",
    "plt.title('Comparison of Model AUCs')\n",
    "plt.ylabel('AUC Value')\n",
    "\n",
    "# Add a legend to the plot\n",
    "plt.legend()\n",
    "\n",
    "# Display the plot\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "masterthesis",
   "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": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}