{"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_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"\"Kaggle\"","metadata":{}},{"cell_type":"markdown","source":"## Data","metadata":{"papermill":{"duration":0.004271,"end_time":"2023-01-03T13:39:57.223551","exception":false,"start_time":"2023-01-03T13:39:57.21928","status":"completed"},"tags":[]}},{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nimport os\nimport torch\nimport torch.nn as nn\nimport pandas as pd\nfrom torchvision import transforms\nfrom torchvision.io import read_image\nfrom torch.utils.data import DataLoader\nfrom torch.utils.data import Dataset\nfrom PIL import Image\nfrom typing import Dict, List, Tuple\nfrom tqdm import tqdm\nimport pandas as pd","metadata":{"papermill":{"duration":3.412991,"end_time":"2023-01-03T13:40:00.640786","exception":false,"start_time":"2023-01-03T13:39:57.227795","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:49.849330Z","iopub.execute_input":"2023-01-08T04:24:49.849714Z","iopub.status.idle":"2023-01-08T04:24:49.858095Z","shell.execute_reply.started":"2023-01-08T04:24:49.849676Z","shell.execute_reply":"2023-01-08T04:24:49.857005Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import sys\nsys.path.append(\"/kaggle/input/timm-pytorch-image-models/pytorch-image-models-master\")\nimport timm\n\ntimm.list_models(\"efficient*\", pretrained=True)","metadata":{"execution":{"iopub.status.busy":"2023-01-08T04:24:49.860234Z","iopub.execute_input":"2023-01-08T04:24:49.860895Z","iopub.status.idle":"2023-01-08T04:24:49.871668Z","shell.execute_reply.started":"2023-01-08T04:24:49.860840Z","shell.execute_reply":"2023-01-08T04:24:49.870616Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def class2dict(f) -> Dict:\n ans = dict()\n for name in dir(f):\n if name.startswith(\"__\"):\n continue\n if not _is_primitive(getattr(f, name)):\n ans[name] = class2dict(getattr(f, name))\n else:\n ans[name] = getattr(f, name)\n return ans\n\n\ndef _is_primitive(value) -> bool:\n primitive = (int, str, bool, float, List, Dict)\n return type(value) in primitive","metadata":{"execution":{"iopub.status.busy":"2023-01-08T04:24:49.873338Z","iopub.execute_input":"2023-01-08T04:24:49.874080Z","iopub.status.idle":"2023-01-08T04:24:49.881224Z","shell.execute_reply.started":"2023-01-08T04:24:49.874044Z","shell.execute_reply":"2023-01-08T04:24:49.880339Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Config","metadata":{}},{"cell_type":"code","source":"class Config: \n batch_size = 64\n num_workers = 8\n n_epochs = 15\n lr = 1e-5\n model_name = \"vit_large_patch16_224\"\n is_kaggle_notebook = True\n resized_height = 300 \n resized_width = 600\n \n base_input_path_for_kaggle = \"/kaggle/input/cassava-leaf-disease-classification\"\n base_input_path_for_local = \"./\"\n \n @property\n def base_input_path(self):\n return self.base_input_path_for_kaggle if self.is_kaggle_notebook else self.base_input_path_for_local","metadata":{"execution":{"iopub.status.busy":"2023-01-08T04:24:49.884265Z","iopub.execute_input":"2023-01-08T04:24:49.884943Z","iopub.status.idle":"2023-01-08T04:24:49.891385Z","shell.execute_reply.started":"2023-01-08T04:24:49.884907Z","shell.execute_reply":"2023-01-08T04:24:49.890423Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"config = Config()\nconfig.n_epochs = 15\nconfig.batch_size = 64\nconfig.num_workers = 8\nconfig.lr = 5e-5\nconfig.model_name = \"efficientnetv2_rw_t\"\nconfig.is_kaggle_notebook = True\nconfig.resized_height = 600\nconfig.resized_width = 800\n\nconfig.base_input_path","metadata":{"papermill":{"duration":0.018484,"end_time":"2023-01-03T13:40:00.664302","exception":false,"start_time":"2023-01-03T13:40:00.645818","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:49.893774Z","iopub.execute_input":"2023-01-08T04:24:49.894531Z","iopub.status.idle":"2023-01-08T04:24:49.905102Z","shell.execute_reply.started":"2023-01-08T04:24:49.894495Z","shell.execute_reply":"2023-01-08T04:24:49.903968Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_path = os.path.join(config.base_input_path, \"train.csv\")\ntrain_img_dir_path = os.path.join(config.base_input_path, \"train_images\")\ntrain_df = pd.read_csv(train_path)\ntrain_df.label.value_counts().plot(kind=\"bar\")","metadata":{"papermill":{"duration":0.33751,"end_time":"2023-01-03T13:40:01.006509","exception":false,"start_time":"2023-01-03T13:40:00.668999","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:49.906697Z","iopub.execute_input":"2023-01-08T04:24:49.907386Z","iopub.status.idle":"2023-01-08T04:24:50.117997Z","shell.execute_reply.started":"2023-01-08T04:24:49.907351Z","shell.execute_reply":"2023-01-08T04:24:50.116783Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from typing import Optional\nfrom torch.utils.data import Dataset\nimport torch.nn as nn\nfrom torchvision.transforms import transforms\n\nclass CassavaDataset(Dataset):\n def __init__(self, annotation_path: Optional[str], img_dir_path: str, img_transforms: transforms.Compose):\n self.has_label = annotation_path is not None\n if annotation_path:\n self.annotation_path = annotation_path\n self.annotation_data = pd.read_csv(annotation_path)\n else:\n self.annotation_path = None\n self.annotation_data = pd.DataFrame()\n self.annotation_data[\"image_id\"] = list(os.listdir(img_dir_path))\n self.img_dir_path = img_dir_path\n self.img_transforms = img_transforms\n \n def __len__(self) -> int:\n return len(self.annotation_data)\n \n def __getitem__(self, index: int):\n data = self.annotation_data.iloc[index, :]\n image_id = data[\"image_id\"]\n if self.has_label:\n label = data[\"label\"]\n image_path = os.path.join(self.img_dir_path, image_id) \n img = Image.open(image_path).convert(\"RGB\")\n if self.img_transforms:\n img = self.img_transforms(img)\n if self.has_label:\n label = torch.nn.functional.one_hot(torch.tensor([label]), num_classes=5)\n label = torch.squeeze(label, dim=0).float()\n if self.has_label:\n return img, image_id, label\n else:\n return img, image_id","metadata":{"papermill":{"duration":0.015811,"end_time":"2023-01-03T13:40:01.027496","exception":false,"start_time":"2023-01-03T13:40:01.011685","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.119425Z","iopub.execute_input":"2023-01-08T04:24:50.119743Z","iopub.status.idle":"2023-01-08T04:24:50.131427Z","shell.execute_reply.started":"2023-01-08T04:24:50.119717Z","shell.execute_reply":"2023-01-08T04:24:50.130276Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Network","metadata":{"papermill":{"duration":0.004436,"end_time":"2023-01-03T13:40:01.03669","exception":false,"start_time":"2023-01-03T13:40:01.032254","status":"completed"},"tags":[]}},{"cell_type":"code","source":"import timm\n\n\nclass CassavaNetwork(nn.Module):\n def __init__(self, config: Config, output_dim: int = 5):\n super().__init__()\n self.base_model = timm.create_model(\n config.model_name,\n pretrained=not config.is_kaggle_notebook,\n num_classes=output_dim,\n )\n \n def forward(self, img: torch.Tensor) -> torch.Tensor:\n return self.base_model(img)\n \n\nmodel = CassavaNetwork(config=config)","metadata":{"papermill":{"duration":1.262427,"end_time":"2023-01-03T13:40:02.304048","exception":false,"start_time":"2023-01-03T13:40:01.041621","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.134218Z","iopub.execute_input":"2023-01-08T04:24:50.135036Z","iopub.status.idle":"2023-01-08T04:24:50.476465Z","shell.execute_reply.started":"2023-01-08T04:24:50.134998Z","shell.execute_reply":"2023-01-08T04:24:50.475357Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Trainer","metadata":{"papermill":{"duration":0.005608,"end_time":"2023-01-03T13:40:02.315177","exception":false,"start_time":"2023-01-03T13:40:02.309569","status":"completed"},"tags":[]}},{"cell_type":"code","source":"import numpy as np\nfrom typing import Optional\nfrom torch.cuda.amp.grad_scaler import GradScaler\nfrom torch.cuda.amp.autocast_mode import autocast\nimport torch.nn as nn\nimport torch.optim as optim\nfrom torch.utils.data import DataLoader\nimport torch\n\nfrom tqdm import tqdm\nimport sys\nimport os\n\n\nclass TrainerOutput:\n loss: float\n\n def __init__(self, loss: float):\n self.loss = loss\n\n @property\n def score(self) -> float:\n return self.loss\n\n\nclass Trainer:\n model: nn.Module\n config: Config\n optimizer: optim.Optimizer\n lr_scheduler: optim.lr_scheduler._LRScheduler\n loss_function: nn.CrossEntropyLoss\n dataloader: DataLoader\n scaler: Optional[GradScaler]\n\n def __init__(\n self,\n model: nn.Module,\n config: Config,\n optimizer: optim.Optimizer,\n lr_scheduler: optim.lr_scheduler._LRScheduler,\n loss_function: nn.CrossEntropyLoss,\n dataloader: DataLoader,\n scaler: Optional[GradScaler] = None,\n ):\n self.model = model\n self.config = config\n self.optimizer = optimizer\n self.lr_scheduler = lr_scheduler\n self.loss_function = loss_function\n self.dataloader = dataloader\n self.scaler = scaler\n\n def advance(self, verbose: bool = False) -> TrainerOutput:\n scaler = None\n if self.scaler:\n scaler = self.scaler\n self.model = self.model.train()\n GPU_DEVICE = torch.device(\"cuda\")\n epoch_loss = 0\n with autocast(enabled=True):\n with torch.enable_grad():\n for batch in tqdm(self.dataloader):\n if len(batch) == 3:\n imgs, ids, labels = batch\n elif len(batch) == 2:\n imgs, labels = batch\n else:\n continue\n self.optimizer.zero_grad()\n imgs = imgs.to(GPU_DEVICE)\n labels = labels.to(GPU_DEVICE)\n out: torch.Tensor = self.model(imgs)\n out = out.float()\n if verbose:\n print(\"out-shape\", out.shape)\n print(\"out\", out) \n if verbose:\n print(\"labels\", labels)\n loss = self.loss_function(out, labels)\n if verbose:\n print(\"loss\", loss)\n if scaler is not None:\n scaler.scale(loss).backward()\n else:\n loss.backward()\n epoch_loss += loss.item()\n if scaler is not None:\n scaler.step(self.optimizer)\n scaler.update()\n else:\n self.optimizer.step()\n self.lr_scheduler.step()\n loss = epoch_loss / (len(self.dataloader))\n return TrainerOutput(loss=loss)\n \n def save_model(self):\n config = self.config\n out_path = os.path.join(\"./\", \"weights\")\n os.makedirs(out_path, exist_ok=True)\n out_path = os.path.join(out_path, f\"epoch_{config.n_epochs}_base_{config.model_name}.pth\")\n torch.save(self.model.state_dict(), out_path)","metadata":{"papermill":{"duration":0.024654,"end_time":"2023-01-03T13:40:02.34482","exception":false,"start_time":"2023-01-03T13:40:02.320166","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.480588Z","iopub.execute_input":"2023-01-08T04:24:50.481031Z","iopub.status.idle":"2023-01-08T04:24:50.498570Z","shell.execute_reply.started":"2023-01-08T04:24:50.480991Z","shell.execute_reply":"2023-01-08T04:24:50.497409Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Classifier","metadata":{"papermill":{"duration":0.004534,"end_time":"2023-01-03T13:40:02.354125","exception":false,"start_time":"2023-01-03T13:40:02.349591","status":"completed"},"tags":[]}},{"cell_type":"code","source":"from typing import Callable, Dict, List, Tuple, Union, Optional\nfrom typing_extensions import Self\nimport numpy as np\nfrom torch import Tensor\nimport torch\nimport torch.nn as nn\nfrom torch.utils.data import DataLoader\n\nimport sys\nimport os\nimport pandas as pd\nfrom tqdm import tqdm\n\nclass ClassifierOutput:\n move_to_full_nn_img_ids: List[str]\n count: int\n acc: float\n loss: float\n df: pd.DataFrame\n\n def __init__(\n self,\n count: int,\n acc: float,\n loss: float,\n df: pd.DataFrame = pd.DataFrame()\n ):\n self.count = count \n self.acc = acc\n self.loss = loss\n self.df = df\n\n @property\n def score(self) -> float:\n return self.acc\n\n\nclass ClassifierTrainInput:\n def __init__(self):\n pass\n\nclass ClassifierTestInput:\n def __init__(self):\n pass\n\nclass Classifier:\n\n model: nn.Module\n activate_function: nn.Softmax\n loss_function: Optional[nn.CrossEntropyLoss]\n c_high: float\n c_low: float\n dataloader: DataLoader\n phase: str\n phase_input: Union[ClassifierTrainInput, ClassifierTestInput]\n on_classify: Optional[Callable]\n\n def __init__(\n self,\n model: nn.Module,\n activate_function: nn.Softmax,\n dataloader: DataLoader,\n phase: str,\n phase_input: Union[ClassifierTrainInput, ClassifierTestInput],\n loss_function: Optional[nn.CrossEntropyLoss] = None,\n on_classify: Optional[Callable] = None,\n ):\n self.model = model\n self.activate_function = activate_function\n self.loss_function = loss_function\n self.dataloader = dataloader\n self.phase = phase\n self.phase_input = phase_input\n self.on_classify = on_classify\n\n def infer(\n self, verbose: bool = False, handle_all: bool = False, calc_acc: bool = True\n ) -> ClassifierOutput:\n net = self.model\n loader = self.dataloader\n loss_function = self.loss_function\n count = 0\n correct = 0 \n GPU_DEVICE = torch.device(\"cuda\")\n called = False\n net = net.eval()\n epoch_loss = 0\n label_df = pd.DataFrame()\n with torch.no_grad():\n for batch in tqdm(loader):\n if len(batch) == 3:\n imgs, ids, labels = batch\n elif len(batch) == 2:\n imgs, ids = batch\n labels = None\n else:\n continue\n imgs = imgs.to(GPU_DEVICE)\n if labels is not None:\n labels = labels.to(GPU_DEVICE) \n outputs = net(imgs)\n if loss_function and labels is not None:\n loss = loss_function(outputs, labels)\n epoch_loss += loss.item()\n outputs = self.activate_function(outputs)\n if verbose:\n print(\"output\", outputs[:10])\n _, pred_indexes = torch.max(outputs, dim=1)\n pred_indexes: Tensor = pred_indexes\n if verbose:\n print(f\"{self.phase}_predicted_indexes\", pred_indexes)\n print(f\"{self.phase}_labels\", labels)\n if not called and self.on_classify is not None:\n if not handle_all:\n called = True\n self.on_classify(imgs, pred_indexes)\n \n if labels is not None:\n labels = torch.tensor(list(map(lambda nums: (nums==1).nonzero().item(), labels))).to(GPU_DEVICE)\n count += pred_indexes.shape[0]\n batch_df = pd.DataFrame()\n batch_df[\"image_id\"] = ids\n batch_df[\"label\"] = pd.Series(pred_indexes.detach().cpu().numpy())\n if len(label_df) == 0:\n label_df = batch_df\n else:\n label_df = pd.concat([label_df, batch_df]) \n if labels is not None and calc_acc:\n correct += int(torch.where(pred_indexes == labels, 1.0, 0.0).sum()) \n acc = correct / count\n epoch_loss = epoch_loss / (len(loader))\n label_df.reset_index(drop=True, inplace=True)\n ret = ClassifierOutput(\n count=count,\n acc=acc,\n loss=epoch_loss,\n df=label_df\n ) \n return ret","metadata":{"papermill":{"duration":0.031072,"end_time":"2023-01-03T13:40:02.389767","exception":false,"start_time":"2023-01-03T13:40:02.358695","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.500204Z","iopub.execute_input":"2023-01-08T04:24:50.500804Z","iopub.status.idle":"2023-01-08T04:24:50.524536Z","shell.execute_reply.started":"2023-01-08T04:24:50.500768Z","shell.execute_reply":"2023-01-08T04:24:50.523597Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Run","metadata":{"papermill":{"duration":0.004264,"end_time":"2023-01-03T13:40:02.398965","exception":false,"start_time":"2023-01-03T13:40:02.394701","status":"completed"},"tags":[]}},{"cell_type":"code","source":"import torch.optim as optim\nfrom sklearn.model_selection import KFold\nfrom torch.utils.data.dataset import Subset\n\nindex = 3\nname = f\"run-{index}\"\nnotes = \"\"\n\ntrain_path = os.path.join(config.base_input_path, \"train.csv\")\ntrain_img_dir_path = os.path.join(config.base_input_path, \"train_images\")\ndf = pd.read_csv(train_path)\n\ndf.label.value_counts().plot(kind=\"bar\")","metadata":{"papermill":{"duration":1.808145,"end_time":"2023-01-03T13:40:04.211843","exception":false,"start_time":"2023-01-03T13:40:02.403698","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.527262Z","iopub.execute_input":"2023-01-08T04:24:50.527521Z","iopub.status.idle":"2023-01-08T04:24:50.738510Z","shell.execute_reply.started":"2023-01-08T04:24:50.527497Z","shell.execute_reply":"2023-01-08T04:24:50.737519Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport torch\nimport os\n\ndef set_seed(seed):\n np.random.seed(seed)\n random_state = np.random.RandomState(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.backends.cudnn.deterministic = True\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n return random_state","metadata":{"execution":{"iopub.status.busy":"2023-01-08T04:24:50.740085Z","iopub.execute_input":"2023-01-08T04:24:50.740697Z","iopub.status.idle":"2023-01-08T04:24:50.748744Z","shell.execute_reply.started":"2023-01-08T04:24:50.740658Z","shell.execute_reply":"2023-01-08T04:24:50.747657Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"set_seed(3407)","metadata":{"papermill":{"duration":0.021793,"end_time":"2023-01-03T13:40:04.239622","exception":false,"start_time":"2023-01-03T13:40:04.217829","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.751090Z","iopub.execute_input":"2023-01-08T04:24:50.751971Z","iopub.status.idle":"2023-01-08T04:24:50.759918Z","shell.execute_reply.started":"2023-01-08T04:24:50.751935Z","shell.execute_reply":"2023-01-08T04:24:50.758596Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"k_fold = KFold(n_splits=2)\ntrain_dataset = CassavaDataset(\n annotation_path=train_path,\n img_dir_path=train_img_dir_path,\n img_transforms=transforms.Compose([\n transforms.ToTensor(), \n transforms.RandomHorizontalFlip(),\n transforms.RandomVerticalFlip(),\n transforms.ColorJitter(brightness=0.5, contrast=0.5, saturation=0.5),\n transforms.Resize((config.resized_height, config.resized_width)),\n transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),\n ]),\n)\nval_dataset = CassavaDataset(\n annotation_path=train_path,\n img_dir_path=train_img_dir_path,\n img_transforms=transforms.Compose([\n transforms.ToTensor(),\n transforms.Resize((config.resized_height, config.resized_width)),\n transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)), \n ]),\n)","metadata":{"execution":{"iopub.status.busy":"2023-01-08T04:24:50.761912Z","iopub.execute_input":"2023-01-08T04:24:50.762568Z","iopub.status.idle":"2023-01-08T04:24:50.793494Z","shell.execute_reply.started":"2023-01-08T04:24:50.762534Z","shell.execute_reply":"2023-01-08T04:24:50.792652Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from torch.cuda.amp.grad_scaler import GradScaler\n\nmodel = CassavaNetwork(config=config)\nmodel = model.to(torch.device(\"cuda\"))\nmodel = nn.DataParallel(model)\n\nif config.is_kaggle_notebook:\n pretrained_weight_path = \"/kaggle/input/weights/epoch_15_base_efficientnetv2_rw_t.pth\"\n model.load_state_dict(torch.load(pretrained_weight_path))\nelse:\n for fold, (train_index, val_index) in enumerate(k_fold.split(df)):\n train_subset = Subset(train_dataset, train_index)\n train_dataloader = DataLoader(\n train_subset,\n batch_size=config.batch_size,\n num_workers=config.num_workers,\n shuffle=True,\n )\n val_subset = Subset(val_dataset, val_index)\n val_dataloader = DataLoader(\n val_subset,\n batch_size=config.batch_size,\n num_workers=config.num_workers,\n shuffle=False,\n ) \n\n optimizer = optim.Adam(model.parameters(), lr=config.lr)\n scheduler = optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.95)\n loss_function = nn.CrossEntropyLoss()\n\n activate_function = nn.Softmax(dim=1)\n \n scaler = GradScaler()\n\n trainer = Trainer(\n model,\n config,\n optimizer,\n scheduler,\n loss_function,\n train_dataloader,\n scaler=scaler,\n )\n classifier = Classifier(\n model,\n activate_function,\n train_dataloader,\n phase=\"train\",\n phase_input=ClassifierTrainInput(),\n )\n\n val_classifier = Classifier(\n model,\n activate_function,\n val_dataloader,\n phase=\"test\",\n phase_input=ClassifierTestInput(),\n )\n\n\n for epoch in tqdm(range(config.n_epochs)):\n epoch += 1\n train_out = trainer.advance()\n print(f\"epoch: {epoch}, train loss: {train_out.loss}\")\n # train_infer_out = classifier.infer()\n # print(f\"epoch: {epoch}, train acc: {train_infer_out.acc}\")\n val_infer_out = val_classifier.infer()\n print(f\"epoch: {epoch}, val acc: {val_infer_out.acc}\")\n if epoch == config.n_epochs:\n trainer.save_model()","metadata":{"papermill":{"duration":7267.6653,"end_time":"2023-01-03T15:41:11.909946","exception":false,"start_time":"2023-01-03T13:40:04.244646","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:50.796446Z","iopub.execute_input":"2023-01-08T04:24:50.796699Z","iopub.status.idle":"2023-01-08T04:24:51.511715Z","shell.execute_reply.started":"2023-01-08T04:24:50.796675Z","shell.execute_reply":"2023-01-08T04:24:51.510773Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Submission","metadata":{"papermill":{"duration":0.847939,"end_time":"2023-01-03T15:41:13.562158","exception":false,"start_time":"2023-01-03T15:41:12.714219","status":"completed"},"tags":[]}},{"cell_type":"code","source":"test_data_img_dir_path = os.path.join(config.base_input_path, \"test_images\")\ntest_dataset = CassavaDataset(\n annotation_path=None,\n img_dir_path=test_data_img_dir_path,\n img_transforms=transforms.Compose([\n transforms.ToTensor(),\n transforms.Resize((config.resized_height, config.resized_width)),\n transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),\n ]),\n)\n\nactivate_function = nn.Softmax(dim=1)\n\ntest_dataloader = DataLoader(\n test_dataset,\n batch_size=config.batch_size,\n num_workers=config.num_workers,\n)\n\ntest_classifier = Classifier(\n model,\n activate_function,\n test_dataloader,\n phase=\"test\",\n phase_input=ClassifierTrainInput(),\n)\n\nout = test_classifier.infer(calc_acc=False, verbose=True)\nout.df.to_csv(\"submission.csv\", index=False)","metadata":{"papermill":{"duration":1.48478,"end_time":"2023-01-03T15:41:15.840399","exception":false,"start_time":"2023-01-03T15:41:14.355619","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-01-08T04:24:51.516184Z","iopub.execute_input":"2023-01-08T04:24:51.518783Z","iopub.status.idle":"2023-01-08T04:24:52.103428Z","shell.execute_reply.started":"2023-01-08T04:24:51.518743Z","shell.execute_reply":"2023-01-08T04:24:52.102344Z"},"trusted":true},"execution_count":null,"outputs":[]}]}