{"cells":[{"metadata":{},"cell_type":"markdown","source":"# About this notebook\n- PyTorch resnext50_32x4d starter code\n- GroupKFold 4 folds\n\nIf this notebook is helpful, feel free to upvote :)"},{"metadata":{"papermill":{"duration":0.021363,"end_time":"2020-12-14T19:52:40.311302","exception":false,"start_time":"2020-12-14T19:52:40.289939","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Data Loading"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:40.356269Z","iopub.status.busy":"2020-12-14T19:52:40.355455Z","iopub.status.idle":"2020-12-14T19:52:41.346426Z","shell.execute_reply":"2020-12-14T19:52:41.344656Z"},"papermill":{"duration":1.015612,"end_time":"2020-12-14T19:52:41.346578","exception":false,"start_time":"2020-12-14T19:52:40.330966","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"import os\n\nimport pandas as pd\n\nfrom matplotlib import pyplot as plt\nimport seaborn as sns","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"os.listdir('../input/ranzcr-clip-catheter-line-classification')","execution_count":null,"outputs":[]},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:41.396141Z","iopub.status.busy":"2020-12-14T19:52:41.395396Z","iopub.status.idle":"2020-12-14T19:52:41.45231Z","shell.execute_reply":"2020-12-14T19:52:41.451677Z"},"papermill":{"duration":0.084553,"end_time":"2020-12-14T19:52:41.45243","exception":false,"start_time":"2020-12-14T19:52:41.367877","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"train = pd.read_csv('../input/ranzcr-clip-catheter-line-classification/train.csv')\ntest = pd.read_csv('../input/ranzcr-clip-catheter-line-classification/sample_submission.csv')\ndisplay(train.head())\ndisplay(test.head())","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Quick EDA"},{"metadata":{"trusted":true},"cell_type":"code","source":"train['PatientID'].value_counts()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"target_cols = ['ETT - Abnormal', 'ETT - Borderline', 'ETT - Normal', 'NGT - Abnormal', \n               'NGT - Borderline', 'NGT - Incompletely Imaged', 'NGT - Normal', 'CVC - Abnormal',\n               'CVC - Borderline', 'CVC - Normal', 'Swan Ganz Catheter Present']\nfor c in target_cols:\n    plt.hist(train[c].values)\n    plt.title(f'target: {c}')\n    plt.show()","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.022224,"end_time":"2020-12-14T19:52:41.498163","exception":false,"start_time":"2020-12-14T19:52:41.475939","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Directory settings"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:41.549032Z","iopub.status.busy":"2020-12-14T19:52:41.548343Z","iopub.status.idle":"2020-12-14T19:52:41.553026Z","shell.execute_reply":"2020-12-14T19:52:41.552269Z"},"papermill":{"duration":0.033458,"end_time":"2020-12-14T19:52:41.553131","exception":false,"start_time":"2020-12-14T19:52:41.519673","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Directory settings\n# ====================================================\nimport os\n\nOUTPUT_DIR = './'\nif not os.path.exists(OUTPUT_DIR):\n    os.makedirs(OUTPUT_DIR)\n\n#TRAIN_PATH = '../input/ranzcr-512x512-dataset'\nTRAIN_PATH = '../input/ranzcr-640x640-dataset'","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.02073,"end_time":"2020-12-14T19:52:41.594446","exception":false,"start_time":"2020-12-14T19:52:41.573716","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# CFG"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:41.647322Z","iopub.status.busy":"2020-12-14T19:52:41.646482Z","iopub.status.idle":"2020-12-14T19:52:41.650102Z","shell.execute_reply":"2020-12-14T19:52:41.649535Z"},"papermill":{"duration":0.034829,"end_time":"2020-12-14T19:52:41.650216","exception":false,"start_time":"2020-12-14T19:52:41.615387","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# CFG\n# ====================================================\nclass CFG:\n    debug=False\n    print_freq=100\n    num_workers=4\n    model_name='resnext50_32x4d'\n    size=600\n    scheduler='CosineAnnealingLR' # ['ReduceLROnPlateau', 'CosineAnnealingLR', 'CosineAnnealingWarmRestarts']\n    epochs=6\n    #factor=0.2 # ReduceLROnPlateau\n    #patience=4 # ReduceLROnPlateau\n    #eps=1e-6 # ReduceLROnPlateau\n    T_max=6 # CosineAnnealingLR\n    #T_0=6 # CosineAnnealingWarmRestarts\n    lr=1e-4\n    min_lr=1e-6\n    batch_size=32\n    weight_decay=1e-6\n    gradient_accumulation_steps=1\n    max_grad_norm=1000\n    seed=42\n    target_size=11\n    target_cols=['ETT - Abnormal', 'ETT - Borderline', 'ETT - Normal',\n                 'NGT - Abnormal', 'NGT - Borderline', 'NGT - Incompletely Imaged', 'NGT - Normal', \n                 'CVC - Abnormal', 'CVC - Borderline', 'CVC - Normal',\n                 'Swan Ganz Catheter Present']\n    n_fold=4\n    trn_fold=[0, 1, 2, 3]\n    train=True\n    \nif CFG.debug:\n    CFG.epochs = 1\n    train = train.sample(n=100, random_state=CFG.seed).reset_index(drop=True)","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.02152,"end_time":"2020-12-14T19:52:41.693202","exception":false,"start_time":"2020-12-14T19:52:41.671682","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Library"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:41.750878Z","iopub.status.busy":"2020-12-14T19:52:41.750245Z","iopub.status.idle":"2020-12-14T19:52:45.494184Z","shell.execute_reply":"2020-12-14T19:52:45.492665Z"},"papermill":{"duration":3.779959,"end_time":"2020-12-14T19:52:45.49431","exception":false,"start_time":"2020-12-14T19:52:41.714351","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Library\n# ====================================================\nimport sys\nsys.path.append('../input/pytorch-image-models/pytorch-image-models-master')\n\nimport os\nimport math\nimport time\nimport random\nimport shutil\nfrom pathlib import Path\nfrom contextlib import contextmanager\nfrom collections import defaultdict, Counter\n\nimport scipy as sp\nimport numpy as np\nimport pandas as pd\n\nfrom sklearn import preprocessing\nfrom sklearn.metrics import roc_auc_score\nfrom sklearn.model_selection import StratifiedKFold, GroupKFold, KFold\n\nfrom tqdm.auto import tqdm\nfrom functools import partial\n\nimport cv2\nfrom PIL import Image\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.optim import Adam, SGD\nimport torchvision.models as models\nfrom torch.nn.parameter import Parameter\nfrom torch.utils.data import DataLoader, Dataset\nfrom torch.optim.lr_scheduler import CosineAnnealingWarmRestarts, CosineAnnealingLR, ReduceLROnPlateau\n\nfrom albumentations import (\n    Compose, OneOf, Normalize, Resize, RandomResizedCrop, RandomCrop, HorizontalFlip, VerticalFlip, \n    RandomBrightness, RandomContrast, RandomBrightnessContrast, Rotate, ShiftScaleRotate, Cutout, \n    IAAAdditiveGaussianNoise, Transpose\n    )\nfrom albumentations.pytorch import ToTensorV2\nfrom albumentations import ImageOnlyTransform\n\nimport timm\n\nfrom torch.cuda.amp import autocast, GradScaler\n\nimport warnings \nwarnings.filterwarnings('ignore')\n\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.021243,"end_time":"2020-12-14T19:52:45.536479","exception":false,"start_time":"2020-12-14T19:52:45.515236","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Utils"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:45.59513Z","iopub.status.busy":"2020-12-14T19:52:45.594471Z","iopub.status.idle":"2020-12-14T19:52:45.60117Z","shell.execute_reply":"2020-12-14T19:52:45.60042Z"},"papermill":{"duration":0.040687,"end_time":"2020-12-14T19:52:45.601288","exception":false,"start_time":"2020-12-14T19:52:45.560601","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Utils\n# ====================================================\ndef get_score(y_true, y_pred):\n    scores = []\n    for i in range(y_true.shape[1]):\n        score = roc_auc_score(y_true[:,i], y_pred[:,i])\n        scores.append(score)\n    avg_score = np.mean(scores)\n    return avg_score, scores\n\n\n@contextmanager\ndef timer(name):\n    t0 = time.time()\n    LOGGER.info(f'[{name}] start')\n    yield\n    LOGGER.info(f'[{name}] done in {time.time() - t0:.0f} s.')\n\n\ndef init_logger(log_file=OUTPUT_DIR+'train.log'):\n    from logging import getLogger, INFO, FileHandler,  Formatter,  StreamHandler\n    logger = getLogger(__name__)\n    logger.setLevel(INFO)\n    handler1 = StreamHandler()\n    handler1.setFormatter(Formatter(\"%(message)s\"))\n    handler2 = FileHandler(filename=log_file)\n    handler2.setFormatter(Formatter(\"%(message)s\"))\n    logger.addHandler(handler1)\n    logger.addHandler(handler2)\n    return logger\n\nLOGGER = init_logger()\n\n\ndef seed_torch(seed=42):\n    random.seed(seed)\n    os.environ['PYTHONHASHSEED'] = str(seed)\n    np.random.seed(seed)\n    torch.manual_seed(seed)\n    torch.cuda.manual_seed(seed)\n    torch.backends.cudnn.deterministic = True\n\nseed_torch(seed=CFG.seed)","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.020877,"end_time":"2020-12-14T19:52:45.643202","exception":false,"start_time":"2020-12-14T19:52:45.622325","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# CV split"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:45.784606Z","iopub.status.busy":"2020-12-14T19:52:45.783744Z","iopub.status.idle":"2020-12-14T19:52:45.800946Z","shell.execute_reply":"2020-12-14T19:52:45.800261Z"},"papermill":{"duration":0.046951,"end_time":"2020-12-14T19:52:45.80106","exception":false,"start_time":"2020-12-14T19:52:45.754109","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"folds = train.copy()\nFold = GroupKFold(n_splits=CFG.n_fold)\ngroups = folds['PatientID'].values\nfor n, (train_index, val_index) in enumerate(Fold.split(folds, folds[CFG.target_cols], groups)):\n    folds.loc[val_index, 'fold'] = int(n)\nfolds['fold'] = folds['fold'].astype(int)\ndisplay(folds.groupby('fold').size())","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.023234,"end_time":"2020-12-14T19:52:45.910123","exception":false,"start_time":"2020-12-14T19:52:45.886889","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Dataset"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:45.967747Z","iopub.status.busy":"2020-12-14T19:52:45.96573Z","iopub.status.idle":"2020-12-14T19:52:45.968559Z","shell.execute_reply":"2020-12-14T19:52:45.969115Z"},"papermill":{"duration":0.036559,"end_time":"2020-12-14T19:52:45.969232","exception":false,"start_time":"2020-12-14T19:52:45.932673","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Dataset\n# ====================================================\nclass TrainDataset(Dataset):\n    def __init__(self, df, transform=None):\n        self.df = df\n        self.file_names = df['StudyInstanceUID'].values\n        self.labels = df[CFG.target_cols].values\n        self.transform = transform\n        \n    def __len__(self):\n        return len(self.df)\n\n    def __getitem__(self, idx):\n        file_name = self.file_names[idx]\n        file_path = f'{TRAIN_PATH}/{file_name}.png'\n        image = cv2.imread(file_path)\n        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)\n        if self.transform:\n            augmented = self.transform(image=image)\n            image = augmented['image']\n        label = torch.tensor(self.labels[idx]).float()\n        return image, label","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.024022,"end_time":"2020-12-14T19:52:46.01618","exception":false,"start_time":"2020-12-14T19:52:45.992158","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Transforms"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:46.071509Z","iopub.status.busy":"2020-12-14T19:52:46.070642Z","iopub.status.idle":"2020-12-14T19:52:46.074136Z","shell.execute_reply":"2020-12-14T19:52:46.073613Z"},"papermill":{"duration":0.035507,"end_time":"2020-12-14T19:52:46.074245","exception":false,"start_time":"2020-12-14T19:52:46.038738","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Transforms\n# ====================================================\ndef get_transforms(*, data):\n    \n    if data == 'train':\n        return Compose([\n            #Resize(CFG.size, CFG.size),\n            RandomResizedCrop(CFG.size, CFG.size, scale=(0.85, 1.0)),\n            HorizontalFlip(p=0.5),\n            Normalize(\n                mean=[0.485, 0.456, 0.406],\n                std=[0.229, 0.224, 0.225],\n            ),\n            ToTensorV2(),\n        ])\n\n    elif data == 'valid':\n        return Compose([\n            Resize(CFG.size, CFG.size),\n            Normalize(\n                mean=[0.485, 0.456, 0.406],\n                std=[0.229, 0.224, 0.225],\n            ),\n            ToTensorV2(),\n        ])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_dataset = TrainDataset(train, transform=get_transforms(data='train'))\n\nfor i in range(5):\n    image, label = train_dataset[i]\n    plt.imshow(image[0])\n    plt.title(f'label: {label}')\n    plt.show() ","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.022168,"end_time":"2020-12-14T19:52:46.118843","exception":false,"start_time":"2020-12-14T19:52:46.096675","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# MODEL"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:46.17627Z","iopub.status.busy":"2020-12-14T19:52:46.170979Z","iopub.status.idle":"2020-12-14T19:52:46.18312Z","shell.execute_reply":"2020-12-14T19:52:46.183763Z"},"papermill":{"duration":0.042914,"end_time":"2020-12-14T19:52:46.183878","exception":false,"start_time":"2020-12-14T19:52:46.140964","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# MODEL\n# ====================================================\nclass CustomResNext(nn.Module):\n    def __init__(self, model_name='resnext50_32x4d', pretrained=False):\n        super().__init__()\n        self.model = timm.create_model(model_name, pretrained=pretrained)\n        n_features = self.model.fc.in_features\n        self.model.fc = nn.Linear(n_features, CFG.target_size)\n\n    def forward(self, x):\n        x = self.model(x)\n        return x","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.022356,"end_time":"2020-12-14T19:52:46.228883","exception":false,"start_time":"2020-12-14T19:52:46.206527","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Helper functions"},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:46.445205Z","iopub.status.busy":"2020-12-14T19:52:46.443326Z","iopub.status.idle":"2020-12-14T19:52:46.445944Z","shell.execute_reply":"2020-12-14T19:52:46.446449Z"},"papermill":{"duration":0.064625,"end_time":"2020-12-14T19:52:46.446561","exception":false,"start_time":"2020-12-14T19:52:46.381936","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Helper functions\n# ====================================================\nclass AverageMeter(object):\n    \"\"\"Computes and stores the average and current value\"\"\"\n    def __init__(self):\n        self.reset()\n\n    def reset(self):\n        self.val = 0\n        self.avg = 0\n        self.sum = 0\n        self.count = 0\n\n    def update(self, val, n=1):\n        self.val = val\n        self.sum += val * n\n        self.count += n\n        self.avg = self.sum / self.count\n\n\ndef asMinutes(s):\n    m = math.floor(s / 60)\n    s -= m * 60\n    return '%dm %ds' % (m, s)\n\n\ndef timeSince(since, percent):\n    now = time.time()\n    s = now - since\n    es = s / (percent)\n    rs = es - s\n    return '%s (remain %s)' % (asMinutes(s), asMinutes(rs))\n\n\ndef train_fn(train_loader, model, criterion, optimizer, epoch, scheduler, device):\n    scaler = GradScaler()\n    batch_time = AverageMeter()\n    data_time = AverageMeter()\n    losses = AverageMeter()\n    scores = AverageMeter()\n    # switch to train mode\n    model.train()\n    start = end = time.time()\n    global_step = 0\n    for step, (images, labels) in enumerate(train_loader):\n        # measure data loading time\n        data_time.update(time.time() - end)\n        images = images.to(device)\n        labels = labels.to(device)\n        batch_size = labels.size(0)\n        with autocast():\n            y_preds = model(images)\n            loss = criterion(y_preds, labels)\n        # record loss\n        losses.update(loss.item(), batch_size)\n        if CFG.gradient_accumulation_steps > 1:\n            loss = loss / CFG.gradient_accumulation_steps\n        scaler.scale(loss).backward()\n        grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), CFG.max_grad_norm)\n        if (step + 1) % CFG.gradient_accumulation_steps == 0:\n            scaler.step(optimizer)\n            scaler.update()\n            optimizer.zero_grad()\n            global_step += 1\n        # measure elapsed time\n        batch_time.update(time.time() - end)\n        end = time.time()\n        if step % CFG.print_freq == 0 or step == (len(train_loader)-1):\n            print('Epoch: [{0}][{1}/{2}] '\n                  'Data {data_time.val:.3f} ({data_time.avg:.3f}) '\n                  'Elapsed {remain:s} '\n                  'Loss: {loss.val:.4f}({loss.avg:.4f}) '\n                  'Grad: {grad_norm:.4f}  '\n                  #'LR: {lr:.6f}  '\n                  .format(\n                   epoch+1, step, len(train_loader), batch_time=batch_time,\n                   data_time=data_time, loss=losses,\n                   remain=timeSince(start, float(step+1)/len(train_loader)),\n                   grad_norm=grad_norm,\n                   #lr=scheduler.get_lr()[0],\n                   ))\n    return losses.avg\n\n\ndef valid_fn(valid_loader, model, criterion, device):\n    batch_time = AverageMeter()\n    data_time = AverageMeter()\n    losses = AverageMeter()\n    scores = AverageMeter()\n    # switch to evaluation mode\n    model.eval()\n    preds = []\n    start = end = time.time()\n    for step, (images, labels) in enumerate(valid_loader):\n        # measure data loading time\n        data_time.update(time.time() - end)\n        images = images.to(device)\n        labels = labels.to(device)\n        batch_size = labels.size(0)\n        # compute loss\n        with torch.no_grad():\n            y_preds = model(images)\n        loss = criterion(y_preds, labels)\n        losses.update(loss.item(), batch_size)\n        # record accuracy\n        preds.append(y_preds.sigmoid().to('cpu').numpy())\n        if CFG.gradient_accumulation_steps > 1:\n            loss = loss / CFG.gradient_accumulation_steps\n        # measure elapsed time\n        batch_time.update(time.time() - end)\n        end = time.time()\n        if step % CFG.print_freq == 0 or step == (len(valid_loader)-1):\n            print('EVAL: [{0}/{1}] '\n                  'Data {data_time.val:.3f} ({data_time.avg:.3f}) '\n                  'Elapsed {remain:s} '\n                  'Loss: {loss.val:.4f}({loss.avg:.4f}) '\n                  .format(\n                   step, len(valid_loader), batch_time=batch_time,\n                   data_time=data_time, loss=losses,\n                   remain=timeSince(start, float(step+1)/len(valid_loader)),\n                   ))\n    predictions = np.concatenate(preds)\n    return losses.avg, predictions","execution_count":null,"outputs":[]},{"metadata":{"papermill":{"duration":0.022557,"end_time":"2020-12-14T19:52:46.492442","exception":false,"start_time":"2020-12-14T19:52:46.469885","status":"completed"},"tags":[]},"cell_type":"markdown","source":"# Train loop"},{"metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","execution":{"iopub.execute_input":"2020-12-14T19:52:46.552529Z","iopub.status.busy":"2020-12-14T19:52:46.541988Z","iopub.status.idle":"2020-12-14T19:52:46.608359Z","shell.execute_reply":"2020-12-14T19:52:46.609161Z"},"papermill":{"duration":0.093848,"end_time":"2020-12-14T19:52:46.609353","exception":false,"start_time":"2020-12-14T19:52:46.515505","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# Train loop\n# ====================================================\ndef train_loop(folds, fold):\n\n    LOGGER.info(f\"========== fold: {fold} training ==========\")\n\n    # ====================================================\n    # loader\n    # ====================================================\n    trn_idx = folds[folds['fold'] != fold].index\n    val_idx = folds[folds['fold'] == fold].index\n\n    train_folds = folds.loc[trn_idx].reset_index(drop=True)\n    valid_folds = folds.loc[val_idx].reset_index(drop=True)\n    valid_labels = valid_folds[CFG.target_cols].values\n\n    train_dataset = TrainDataset(train_folds, \n                                 transform=get_transforms(data='train'))\n    valid_dataset = TrainDataset(valid_folds, \n                                 transform=get_transforms(data='valid'))\n\n    train_loader = DataLoader(train_dataset, \n                              batch_size=CFG.batch_size, \n                              shuffle=True, \n                              num_workers=CFG.num_workers, pin_memory=True, drop_last=True)\n    valid_loader = DataLoader(valid_dataset, \n                              batch_size=CFG.batch_size * 2, \n                              shuffle=False, \n                              num_workers=CFG.num_workers, pin_memory=True, drop_last=False)\n    \n    # ====================================================\n    # scheduler \n    # ====================================================\n    def get_scheduler(optimizer):\n        if CFG.scheduler=='ReduceLROnPlateau':\n            scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=CFG.factor, patience=CFG.patience, verbose=True, eps=CFG.eps)\n        elif CFG.scheduler=='CosineAnnealingLR':\n            scheduler = CosineAnnealingLR(optimizer, T_max=CFG.T_max, eta_min=CFG.min_lr, last_epoch=-1)\n        elif CFG.scheduler=='CosineAnnealingWarmRestarts':\n            scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=CFG.T_0, T_mult=1, eta_min=CFG.min_lr, last_epoch=-1)\n        return scheduler\n\n    # ====================================================\n    # model & optimizer\n    # ====================================================\n    model = CustomResNext(CFG.model_name, pretrained=True)\n    model.to(device)\n\n    optimizer = Adam(model.parameters(), lr=CFG.lr, weight_decay=CFG.weight_decay, amsgrad=False)\n    scheduler = get_scheduler(optimizer)\n\n    # ====================================================\n    # loop\n    # ====================================================\n    criterion = nn.BCEWithLogitsLoss()\n\n    best_score = 0.\n    best_loss = np.inf\n    \n    for epoch in range(CFG.epochs):\n        \n        start_time = time.time()\n        \n        # train\n        avg_loss = train_fn(train_loader, model, criterion, optimizer, epoch, scheduler, device)\n\n        # eval\n        avg_val_loss, preds = valid_fn(valid_loader, model, criterion, device)\n        \n        if isinstance(scheduler, ReduceLROnPlateau):\n            scheduler.step(avg_val_loss)\n        elif isinstance(scheduler, CosineAnnealingLR):\n            scheduler.step()\n        elif isinstance(scheduler, CosineAnnealingWarmRestarts):\n            scheduler.step()\n\n        # scoring\n        score, scores = get_score(valid_labels, preds)\n\n        elapsed = time.time() - start_time\n\n        LOGGER.info(f'Epoch {epoch+1} - avg_train_loss: {avg_loss:.4f}  avg_val_loss: {avg_val_loss:.4f}  time: {elapsed:.0f}s')\n        LOGGER.info(f'Epoch {epoch+1} - Score: {score:.4f}  Scores: {np.round(scores, decimals=4)}')\n\n        \"\"\"\n        if score > best_score:\n            best_score = score\n            LOGGER.info(f'Epoch {epoch+1} - Save Best Score: {best_score:.4f} Model')\n            torch.save({'model': model.state_dict(), \n                        'preds': preds},\n                        OUTPUT_DIR+f'{CFG.model_name}_fold{fold}_best.pth')\n        \"\"\"\n        \n        if avg_val_loss < best_loss:\n            best_loss = avg_val_loss\n            LOGGER.info(f'Epoch {epoch+1} - Save Best Loss: {best_loss:.4f} Model')\n            torch.save({'model': model.state_dict(), \n                        'preds': preds},\n                        OUTPUT_DIR+f'{CFG.model_name}_fold{fold}_best.pth')\n    \n    check_point = torch.load(OUTPUT_DIR+f'{CFG.model_name}_fold{fold}_best.pth')\n    for c in [f'pred_{c}' for c in CFG.target_cols]:\n        valid_folds[c] = np.nan\n    valid_folds[[f'pred_{c}' for c in CFG.target_cols]] = check_point['preds']\n\n    return valid_folds","execution_count":null,"outputs":[]},{"metadata":{"execution":{"iopub.execute_input":"2020-12-14T19:52:46.697088Z","iopub.status.busy":"2020-12-14T19:52:46.694995Z","iopub.status.idle":"2020-12-14T19:52:46.697997Z","shell.execute_reply":"2020-12-14T19:52:46.698568Z"},"papermill":{"duration":0.043278,"end_time":"2020-12-14T19:52:46.698687","exception":false,"start_time":"2020-12-14T19:52:46.655409","status":"completed"},"tags":[],"trusted":true},"cell_type":"code","source":"# ====================================================\n# main\n# ====================================================\ndef main():\n\n    \"\"\"\n    Prepare: 1.train  2.folds\n    \"\"\"\n\n    def get_result(result_df):\n        preds = result_df[[f'pred_{c}' for c in CFG.target_cols]].values\n        labels = result_df[CFG.target_cols].values\n        score, scores = get_score(labels, preds)\n        LOGGER.info(f'Score: {score:<.4f}  Scores: {np.round(scores, decimals=4)}')\n    \n    if CFG.train:\n        # train \n        oof_df = pd.DataFrame()\n        for fold in range(CFG.n_fold):\n            if fold in CFG.trn_fold:\n                _oof_df = train_loop(folds, fold)\n                oof_df = pd.concat([oof_df, _oof_df])\n                LOGGER.info(f\"========== fold: {fold} result ==========\")\n                get_result(_oof_df)\n        # CV result\n        LOGGER.info(f\"========== CV ==========\")\n        get_result(oof_df)\n        # save result\n        oof_df.to_csv(OUTPUT_DIR+'oof_df.csv', index=False)","execution_count":null,"outputs":[]},{"metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","execution":{"iopub.execute_input":"2020-12-14T19:52:46.757077Z","iopub.status.busy":"2020-12-14T19:52:46.755987Z"},"papermill":{"duration":null,"end_time":null,"exception":false,"start_time":"2020-12-14T19:52:46.725364","status":"running"},"tags":[],"trusted":true},"cell_type":"code","source":"if __name__ == '__main__':\n    main()","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat":4,"nbformat_minor":4}