{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"tez_path = '../input/tez-lib'\neffnet_path = '../input/efficientnet-pytorch/'\nimport sys\nsys.path.append(tez_path)\nsys.path.append(effnet_path)","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"import os\nimport albumentations\nimport pandas as pd\nimport numpy as np\n\nimport tez\nfrom tez.datasets import ImageDataset\n\nimport torch\nimport torch.nn as nn\nfrom torch.nn import functional as F\n\nfrom efficientnet_pytorch import EfficientNet\nfrom tqdm import tqdm","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class LeafModel(nn.Module):\n    def __init__(self, num_classes):\n        super().__init__()\n\n        self.effnet = EfficientNet.from_name(\"efficientnet-b4\")\n        self.dropout = nn.Dropout(0.2)\n        self.out = nn.Linear(1792, num_classes)\n        self.step_scheduler_after = \"epoch\"\n\n    def forward(self, image, targets=None):\n        batch_size, _, _, _ = image.shape\n\n        x = self.effnet.extract_features(image)\n        x = F.adaptive_avg_pool2d(x, 1).reshape(batch_size, -1)\n        outputs = self.out(self.dropout(x))\n        return outputs, None, None\n    \n    def load(self, model_path, device=\"cuda\"):\n        self.device = device\n        if next(self.parameters()).device != self.device:\n            self.to(self.device)\n        model_dict = torch.load(model_path, map_location=device)\n        self.load_state_dict(model_dict[\"state_dict\"])\n        \n    def predict(self, dataset, sampler=None, batch_size=16, n_jobs=1, collate_fn=None):\n        if next(self.parameters()).device != self.device:\n            self.to(self.device)\n\n        if batch_size == 1:\n            n_jobs = 0\n        data_loader = torch.utils.data.DataLoader(\n            dataset, batch_size=batch_size, num_workers=4, sampler=sampler, collate_fn=collate_fn, pin_memory=True\n        )\n\n        if self.training:\n            self.eval()\n\n        final_output = []\n        tk0 = tqdm(data_loader, total=len(data_loader))\n\n        for b_idx, data in enumerate(tk0):\n            with torch.no_grad():\n                out = self.predict_one_step(data)\n                out = out.cpu().detach().numpy()\n                yield out\n\n            tk0.set_postfix(stage=\"test\")\n        tk0.close()\n        \n    def model_fn(self, data):\n        for key, value in data.items():\n            data[key] = value.to(self.device)\n        #if self.fp16:\n        #    with torch.cuda.amp.autocast():\n        #        output, loss, metrics = self(**data)\n        #else:\n        output, loss, metrics = self(**data)\n        return output, loss, metrics\n    \n    def predict_one_step(self, data):\n        output, _, _ = self.model_fn(data)\n        return output","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# augmentations taken from: https://www.kaggle.com/khyeh0719/pytorch-efficientnet-baseline-inference-tta\ntest_aug = albumentations.Compose([\n    albumentations.RandomResizedCrop(512, 512),\n    albumentations.Transpose(p=0.5),\n    albumentations.HorizontalFlip(p=0.5),\n    albumentations.VerticalFlip(p=0.5),\n    albumentations.HueSaturationValue(\n        hue_shift_limit=0.2, \n        sat_shift_limit=0.2,\n        val_shift_limit=0.2, \n        p=0.5\n    ),\n    albumentations.RandomBrightnessContrast(\n        brightness_limit=(-0.1,0.1), \n        contrast_limit=(-0.1, 0.1), \n        p=0.5\n    ),\n    albumentations.Normalize(\n        mean=[0.485, 0.456, 0.406], \n        std=[0.229, 0.224, 0.225], \n        max_pixel_value=255.0, \n        p=1.0\n    )\n], p=1.)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"dfx = pd.read_csv(\"../input/cassava-leaf-disease-classification/sample_submission.csv\")\nimage_path = \"../input/cassava-leaf-disease-classification/test_images/\"\ntest_image_paths = [os.path.join(image_path, x) for x in dfx.image_id.values]\n# fake targets\ntest_targets = dfx.label.values\ntest_dataset = ImageDataset(\n    image_paths=test_image_paths,\n    targets=test_targets,\n    #resize=None,\n    augmentations=test_aug,\n)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_dfx = pd.read_csv(\"../input/cassava-leaf-disease-classification/train.csv\")\nmodel0 = LeafModel(num_classes=train_dfx.label.nunique())\nmodel0.load(\"../input/single-model/efficentnet_model_fold0.bin\", device='cuda')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# run inference 5 times with 5 different model\n\nmodel_list = [model0]\n\ndef run_inference(model):\n    final_preds = None\n    for j in range(20):\n        preds = model.predict(test_dataset, batch_size=64, n_jobs=-1)\n        temp_preds = None\n        for p in preds:\n            if temp_preds is None:\n                temp_preds = p\n            else:\n                temp_preds = np.vstack((temp_preds, p))\n        if final_preds is None:\n            final_preds = temp_preds\n        else:\n            final_preds += temp_preds\n    final_preds /= 20\n    final_preds = final_preds.argmax(axis=1)\n    return final_preds","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"dfx.label = run_inference(model_list[0])\ndfx.to_csv(\"submission.csv\", index=False)","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}