{"cells":[{"metadata":{},"cell_type":"markdown","source":"# Directory settings"},{"metadata":{"trusted":true},"cell_type":"code","source":"import json\nwith open('../input/train-weights-optimization/best_weights.json', 'r') as f:\n    weights_dict = json.load(f)\nweights_dict","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Library"},{"metadata":{"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 glob\nimport shutil\nfrom pathlib import Path\nfrom contextlib import contextmanager\nfrom collections import defaultdict, Counter\nimport yaml\n\nimport scipy as sp\nimport numpy as np\nimport pandas as pd\n\nfrom sklearn import preprocessing\nfrom sklearn.metrics import accuracy_score\nfrom sklearn.model_selection import StratifiedKFold\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, CenterCrop\n    )\nfrom albumentations.pytorch import ToTensorV2\nfrom albumentations import ImageOnlyTransform\n\nimport timm\n\nimport warnings \nwarnings.filterwarnings('ignore')\n\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# CFG"},{"metadata":{"trusted":true},"cell_type":"code","source":"# # ====================================================\n# # CFG\n# # ====================================================\n\nnormal_configs = []\ntta_configs = []\nnormal_model_dirs = []\ntta_model_dirs = []\n\nfor model_dir in weights_dict.keys():\n    assert len(glob.glob(f'{model_dir}/*.yml'))==1\n    config_path = glob.glob(f'{model_dir}/*.yml')[0]\n    with open(config_path) as f:\n        config = yaml.load(f)\n    if 'valid_augmentation' in config['tag'].keys():\n        tta_model_dirs.append(model_dir)\n        tta_configs.append(config)\n    else:\n        normal_model_dirs.append(model_dir)\n        normal_configs.append(config)\n\n\nTRAIN_PATH = '../input/cassava-leaf-disease-classification/train_images'\nTEST_PATH = '../input/cassava-leaf-disease-classification/test_images'\nOUTPUT_DIR = './'","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Utils"},{"metadata":{"trusted":true},"cell_type":"code","source":"# ====================================================\n# Utils\n# ====================================================\ndef get_score(y_true, y_pred):\n    return accuracy_score(y_true, y_pred)\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\n# def init_logger(log_file=OUTPUT_DIR+'inference.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\n#LOGGER = 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\n# seed_torch(seed=CFG['seed'])","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Data Loading"},{"metadata":{"trusted":true},"cell_type":"code","source":"test = pd.read_csv('../input/cassava-leaf-disease-classification/sample_submission.csv')\ntest.head()","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Dataset"},{"metadata":{"trusted":true},"cell_type":"code","source":"# ====================================================\n# Dataset\n# ====================================================\nclass TestDataset(Dataset):\n    def __init__(self, df, transform=None):\n        self.df = df\n        self.file_names = df['image_id'].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'{TEST_PATH}/{file_name}'\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        return image\n    \n    \nclass TTADataset(Dataset):\n    def __init__(self, df, image_path, ttas):\n        self.df = df\n        self.file_names = df['image_id'].values\n        self.labels = df['label'].values\n        self.image_path = image_path\n        self.ttas = ttas\n\n    def __len__(self) -> int:\n        return len(self.df)\n\n    def __getitem__(self, idx):\n        file_name = self.file_names[idx]\n        file_path = f'{self.image_path}/{file_name}'\n        image = cv2.imread(file_path)\n        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)\n\n        imglist=[tta(image=image)['image'] for tta in self.ttas]  # update\n\n        image=torch.stack(imglist)\n        label = torch.tensor(self.labels[idx]).long()\n        \n        return image, label","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Transforms"},{"metadata":{"trusted":true},"cell_type":"code","source":"def _get_augmentations(aug_list, cfg):\n    process = []\n    for aug in aug_list:\n        if aug ==  'Resize':\n            process.append(Resize(cfg['size'], cfg['size']))\n        elif aug == 'RandomResizedCrop':\n            process.append(RandomResizedCrop(cfg['size'], cfg['size']))\n        elif aug == 'CenterCrop':\n            process.append(CenterCrop(CFG['size'], CFG['size']))\n        elif aug == 'Transpose':\n            process.append(Transpose(p=0.5))\n        elif aug == 'HorizontalFlip':\n            process.append(HorizontalFlip(p=0.5))\n        elif aug == 'VerticalFlip':\n            process.append(VerticalFlip(p=0.5))\n        elif aug == 'ShiftScaleRotate':\n            process.append(ShiftScaleRotate(p=0.5))\n        elif aug == 'Normalize':\n            process.append(Normalize(\n                            mean=[0.485, 0.456, 0.406],\n                            std=[0.229, 0.224, 0.225],\n                        ))\n        else:\n            raise ValueError(f'{aug} is not suitable')\n\n    process.append(ToTensorV2())\n\n    return process","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# ====================================================\n# Transforms\n# ====================================================\ndef get_transforms(*, aug_list, cfg):\n    \n    return Compose(\n        _get_augmentations(aug_list, cfg)\n    )","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_ttas(cfg):\n    norm_mean = [0.485, 0.456, 0.406]\n    norm_std = [0.229, 0.224, 0.225]\n\n    oneof_augs = [\n        CenterCrop(cfg['size'], cfg['size']), \n        Resize(cfg['size'], cfg['size'])\n    ]\n\n    ttas = [[\n        Compose([\n            oneof_aug,\n            Normalize(mean=norm_mean, std=norm_std, p=1.),\n            ToTensorV2()\n        ]),\n        Compose([\n            oneof_aug,\n            Transpose(p=1),\n            Normalize(mean=norm_mean, std=norm_std, p=1.),\n            ToTensorV2()\n        ])\n    ] for oneof_aug in oneof_augs]\n\n    # 平滑化\n    ttas = sum(ttas, [])\n    \n    return ttas","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# MODEL"},{"metadata":{"trusted":true},"cell_type":"code","source":"# ====================================================\n# MODEL\n# ====================================================\nclass CustomModel(nn.Module):\n    def __init__(self, model_name, target_size, pretrained=False):\n        super().__init__()\n        self.model = timm.create_model(model_name, pretrained=pretrained)\n        if hasattr(self.model, 'classifier'):\n            n_features = self.model.classifier.in_features\n            self.model.classifier = nn.Linear(n_features, target_size)\n        elif hasattr(self.model, 'fc'):\n            n_features = self.model.fc.in_features\n            self.model.fc = nn.Linear(n_features, target_size)\n        elif hasattr(self.model, 'head'):\n            n_features = self.model.head.in_features\n            self.model.head = nn.Linear(n_features, target_size)\n\n    def forward(self, x):\n        x = self.model(x)\n        return x","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Helper functions"},{"metadata":{"trusted":true},"cell_type":"code","source":"# ====================================================\n# Helper functions\n# ====================================================\ndef inference_normal(model, states, test_loader, device):\n    model.to(device)\n    tk0 = tqdm(enumerate(test_loader), total=len(test_loader))\n    probs = []\n    for i, (images) in tk0:\n        images = images.to(device)\n        avg_preds = []\n        for state in states:\n            model.load_state_dict(state['model'])\n            model.eval()\n            with torch.no_grad():\n                y_preds = model(images)\n            avg_preds.append(y_preds.softmax(1).to('cpu').numpy())\n        avg_preds = np.mean(avg_preds, axis=0)\n        probs.append(avg_preds)\n    probs = np.concatenate(probs)\n    return probs\n\n\ndef inference_tta(model, states, tta_loader, device):\n    model.to(device)\n    tk0 = tqdm(enumerate(tta_loader), total=len(tta_loader))\n    probs = []\n    for i, (images, _) in tk0:\n        images = images.to(device)\n        batch_size, n_crops, c, h, w = images.size()\n        images = images.view(-1, c, h, w)\n        \n        avg_preds = []\n        for state in states:\n            model.load_state_dict(state['model'])\n            model.eval()\n            with torch.no_grad():\n                y_preds = model(images).softmax(1)\n                y_preds = y_preds.view(batch_size, n_crops,-1)\n            avg_preds.append(y_preds.to('cpu').numpy())\n        avg_preds = np.mean(avg_preds, axis=0)\n        probs.append(avg_preds)\n        del images, _, y_preds, avg_preds\n        torch.cuda.empty_cache()\n    probs = np.concatenate(probs)\n    return probs.mean(1)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# inference"},{"metadata":{"trusted":true},"cell_type":"code","source":"def main(config, model_dir):\n    # ====================================================\n    # inference\n    # ====================================================\n    \n    INFO = config['info']\n    TAG = config['tag']\n    CFG = config['cfg']\n    CFG['train'] = False\n    CFG['inference'] = True\n    inference_batch_size = 64\n    \n    seed_torch(seed=CFG['seed'])\n\n    model = CustomModel(TAG['model_name'], CFG['target_size'], pretrained=False)\n    states = [torch.load(path) for path in glob.glob(f'{model_dir}/*.pth')]\n    test_dataset = TestDataset(test, transform=get_transforms(aug_list=['Resize', 'Normalize'], cfg=CFG))\n    test_loader = DataLoader(test_dataset, batch_size=inference_batch_size, shuffle=False, \n                             num_workers=CFG['num_workers'], pin_memory=True)\n    predictions = inference_normal(model, states, test_loader, device)\n    \n    return predictions\n\n\n\ndef main_tta(config, model_dir):\n    # ====================================================\n    # inference\n    # ====================================================\n    \n    INFO = config['info']\n    TAG = config['tag']\n    CFG = config['cfg']\n    CFG['train'] = False\n    CFG['inference'] = True\n    inference_batch_size = 8\n    \n    seed_torch(seed=CFG['seed'])\n\n    model = CustomModel(TAG['model_name'], CFG['target_size'], pretrained=False)\n    states = [torch.load(path) for path in glob.glob(f'{model_dir}/*.pth')]\n    ttas = get_ttas(CFG)\n    tta_dataset = TTADataset(test, TEST_PATH, ttas=ttas)\n    tta_loader = DataLoader(tta_dataset, batch_size=inference_batch_size, shuffle=False, \n                             num_workers=2, pin_memory=True)\n    predictions = inference_tta(model, states, tta_loader, device)\n    \n    return predictions","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# ensemble\npredictions_list = []\nmodel_dir_list = []\nfor config, model_dir in zip(normal_configs, normal_model_dirs):\n    predictions_list.append(main(config, model_dir))\n    model_dir_list.append(model_dir)\n    \nfor config, model_dir in zip(tta_configs, tta_model_dirs):\n    predictions_list.append(main_tta(config, model_dir))\n    model_dir_list.append(model_dir)\n\n\n# predictions = (predictions_list[0] + predictions_list[1]) / 2\n# # submission\n# test['label'] = predictions.argmax(1)\n# test[['image_id', 'label']].to_csv(OUTPUT_DIR+'submission.csv', index=False)\n# test.head()","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# submit"},{"metadata":{"trusted":true},"cell_type":"code","source":"# ensemble\npredictions = np.zeros(predictions_list[0].shape, dtype=predictions_list[0].dtype)\nfor i, key in zip(range(len(predictions_list)), model_dir_list):\n    predictions += predictions_list[i] * weights_dict[key]\n# submission\ntest['label'] = predictions.argmax(1)\ntest[['image_id', 'label']].to_csv(OUTPUT_DIR+'submission.csv', index=False)\ntest.head()","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}