if name == 'main':
\nseed_everything(CFG['seed'])\n\nfolds = StratifiedKFold(n_splits=CFG['fold_num']).split(np.arange(train.shape[0]), train.label.values)\n\nfor fold, (trn_idx, val_idx) in enumerate(folds):\n\n if fold > 0:\n break \n\n print('Inference fold {} started'.format(fold))\n\n valid_ = train.loc[val_idx,:].reset_index(drop=True)\n valid_ds = CassavaDataset(valid_, '../input/cassava-leaf-disease-classification/train_images/', transforms=get_inference_transforms(), output_label=False)\n\n test = pd.DataFrame()\n test['image_id'] = list(os.listdir('../input/cassava-leaf-disease-classification/test_images/'))\n test_ds = CassavaDataset(test, '../input/cassava-leaf-disease-classification/test_images/', transforms=get_inference_transforms(), output_label=False)\n\n val_loader = torch.utils.data.DataLoader(\n valid_ds, \n batch_size=CFG['valid_bs'],\n num_workers=CFG['num_workers'],\n shuffle=False,\n pin_memory=False,\n )\n\n tst_loader = torch.utils.data.DataLoader(\n test_ds, \n batch_size=CFG['valid_bs'],\n num_workers=CFG['num_workers'],\n shuffle=False,\n pin_memory=False,\n )\n\n device = torch.device(CFG['device'])\n model = CassvaImgClassifier(CFG['model_arch'], train.label.nunique()).to(device)\n\n val_preds = []\n tst_preds = []\n\n #for epoch in range(CFG['epochs']-3):\n for i, epoch in enumerate(CFG['used_epochs']): \n model.load_state_dict(torch.load('../input/project-cassava-leaf-disease/{}_fold_{}_{}'.format(CFG['model_arch'], fold, epoch)))\n\n with torch.no_grad():\n for _ in range(CFG['tta']):\n val_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, val_loader, device)]\n tst_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, tst_loader, device)]\n\n val_preds = np.mean(val_preds, axis=0) \n tst_preds = np.mean(tst_preds, axis=0) \n\n print('fold {} validation loss = {:.5f}'.format(fold, log_loss(valid_.label.values, val_preds)))\n print('fold {} validation accuracy = {:.5f}'.format(fold, (valid_.label.values==np.argmax(val_preds, axis=1)).mean()))\n\n del model\n torch.cuda.empty_cache()\n test['label'] = np.argmax(tst_preds, axis=1)\n test.head()\n test.to_csv('submission.csv', index=False)\n print(test)\nif name == 'main':
\nseed_everything(CFG['seed'])\n\nfolds = StratifiedKFold(n_splits=CFG['fold_num']).split(np.arange(train.shape[0]), train.label.values)\n\nfor fold, (trn_idx, val_idx) in enumerate(folds):\n\n if fold > 0:\n break \n\n print('Inference fold {} started'.format(fold))\n\n valid_ = train.loc[val_idx,:].reset_index(drop=True)\n valid_ds = CassavaDataset(valid_, '../input/cassava-leaf-disease-classification/train_images/', transforms=get_inference_transforms(), output_label=False)\n\n test = pd.DataFrame()\n test['image_id'] = list(os.listdir('../input/cassava-leaf-disease-classification/test_images/'))\n test_ds = CassavaDataset(test, '../input/cassava-leaf-disease-classification/test_images/', transforms=get_inference_transforms(), output_label=False)\n\n val_loader = torch.utils.data.DataLoader(\n valid_ds, \n batch_size=CFG['valid_bs'],\n num_workers=CFG['num_workers'],\n shuffle=False,\n pin_memory=False,\n )\n\n tst_loader = torch.utils.data.DataLoader(\n test_ds, \n batch_size=CFG['valid_bs'],\n num_workers=CFG['num_workers'],\n shuffle=False,\n pin_memory=False,\n )\n\n device = torch.device(CFG['device'])\n model = CassvaImgClassifier(CFG['model_arch'], train.label.nunique()).to(device)\n\n val_preds = []\n tst_preds = []\n\n #for epoch in range(CFG['epochs']-3):\n for i, epoch in enumerate(CFG['used_epochs']): \n model.load_state_dict(torch.load('../input/project-cassava-leaf-disease/{}_fold_{}_{}'.format(CFG['model_arch'], fold, epoch)))\n\n with torch.no_grad():\n for _ in range(CFG['tta']):\n val_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, val_loader, device)]\n tst_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, tst_loader, device)]\n\n val_preds = np.mean(val_preds, axis=0) \n tst_preds = np.mean(tst_preds, axis=0) \n\n print('fold {} validation loss = {:.5f}'.format(fold, log_loss(valid_.label.values, val_preds)))\n print('fold {} validation accuracy = {:.5f}'.format(fold, (valid_.label.values==np.argmax(val_preds, axis=1)).mean()))\n\n del model\n torch.cuda.empty_cache()\n test['label'] = np.argmax(tst_preds, axis=1)\n test.head()\n test.to_csv('submission.csv', index=False)\n print(test)\nThis might help: https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215962
", "rawMarkdown": "This might help: https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215962", "votes": null }, { "id": "1198266", "postDate": "02/12/2021 20:59:32", "content": "Thanks for your reply but can you have a look at my code and try to figure out what should I do any suggestions will be helpful.
", "rawMarkdown": "Thanks for your reply but can you have a look at my code and try to figure out what should I do any suggestions will be helpful.", "votes": null }, { "id": "1198270", "postDate": "02/12/2021 21:04:02", "content": "Hard to debug without knowing what inference_one_epoch is. If you look at the recommendation in the post, you can try scoring the train dataset using your logic and see if you get a submission.csv. More than likely, there is an error in the code and it doesnt produce that file.
", "rawMarkdown": "Hard to debug without knowing what inference_one_epoch is. If you look at the recommendation in the post, you can try scoring the train dataset using your logic and see if you get a submission.csv. More than likely, there is an error in the code and it doesnt produce that file.", "votes": null }, { "id": "1198769", "postDate": "02/13/2021 09:34:33", "content": "def inference_one_epoch(model, data_loader, device):
\n model.eval()
image_preds_all = []\n\npbar = tqdm(enumerate(data_loader), total=len(data_loader))\nfor step, (imgs) in pbar:\n imgs = imgs.to(device).float()\n\n image_preds = model(imgs) #output = model(input)\n image_preds_all += [torch.softmax(image_preds, 1).detach().cpu().numpy()]\n\n\nimage_preds_all = np.concatenate(image_preds_all, axis=0)\nreturn image_preds_all\nthis is inference_one_epoch
", "rawMarkdown": "def inference_one_epoch(model, data_loader, device):\n model.eval()\n\n image_preds_all = []\n \n pbar = tqdm(enumerate(data_loader), total=len(data_loader))\n for step, (imgs) in pbar:\n imgs = imgs.to(device).float()\n \n image_preds = model(imgs) #output = model(input)\n image_preds_all += [torch.softmax(image_preds, 1).detach().cpu().numpy()]\n \n \n image_preds_all = np.concatenate(image_preds_all, axis=0)\n return image_preds_all\nthis is inference_one_epoch", "votes": null }, { "id": "1199363", "postDate": "02/13/2021 19:09:20", "content": "did you try the recommendation?
", "rawMarkdown": "did you try the recommendation?", "votes": null }, { "id": "1200330", "postDate": "02/14/2021 15:28:06", "content": "Yes I changed the test path with train path and it produces a 21000 pedictions.
", "rawMarkdown": "Yes I changed the test path with train path and it produces a 21000 pedictions.", "votes": null } ], "comments": [ { "id": 1198162, "author_name": "trushk", "author_url": "", "post_date": "02/12/2021 19:10:21", "content": "This might help: https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215962
", "votes": null, "replies": [ { "id": 1198266, "author_name": "riyajm", "author_url": "", "post_date": "02/12/2021 20:59:32", "content": "Thanks for your reply but can you have a look at my code and try to figure out what should I do any suggestions will be helpful.
", "votes": null, "replies": [] }, { "id": 1198270, "author_name": "trushk", "author_url": "", "post_date": "02/12/2021 21:04:02", "content": "Hard to debug without knowing what inference_one_epoch is. If you look at the recommendation in the post, you can try scoring the train dataset using your logic and see if you get a submission.csv. More than likely, there is an error in the code and it doesnt produce that file.
", "votes": null, "replies": [] }, { "id": 1198769, "author_name": "riyajm", "author_url": "", "post_date": "02/13/2021 09:34:33", "content": "def inference_one_epoch(model, data_loader, device):
\n model.eval()
image_preds_all = []\n\npbar = tqdm(enumerate(data_loader), total=len(data_loader))\nfor step, (imgs) in pbar:\n imgs = imgs.to(device).float()\n\n image_preds = model(imgs) #output = model(input)\n image_preds_all += [torch.softmax(image_preds, 1).detach().cpu().numpy()]\n\n\nimage_preds_all = np.concatenate(image_preds_all, axis=0)\nreturn image_preds_all\nthis is inference_one_epoch
", "votes": null, "replies": [] }, { "id": 1199363, "author_name": "trushk", "author_url": "", "post_date": "02/13/2021 19:09:20", "content": "did you try the recommendation?
", "votes": null, "replies": [] }, { "id": 1200330, "author_name": "riyajm", "author_url": "", "post_date": "02/14/2021 15:28:06", "content": "Yes I changed the test path with train path and it produces a 21000 pedictions.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1198114": "if __name__ == '__main__':\n \n\n seed_everything(CFG['seed'])\n \n folds = StratifiedKFold(n_splits=CFG['fold_num']).split(np.arange(train.shape[0]), train.label.values)\n \n for fold, (trn_idx, val_idx) in enumerate(folds):\n \n if fold > 0:\n break \n\n print('Inference fold {} started'.format(fold))\n\n valid_ = train.loc[val_idx,:].reset_index(drop=True)\n valid_ds = CassavaDataset(valid_, '../input/cassava-leaf-disease-classification/train_images/', transforms=get_inference_transforms(), output_label=False)\n \n test = pd.DataFrame()\n test['image_id'] = list(os.listdir('../input/cassava-leaf-disease-classification/test_images/'))\n test_ds = CassavaDataset(test, '../input/cassava-leaf-disease-classification/test_images/', transforms=get_inference_transforms(), output_label=False)\n \n val_loader = torch.utils.data.DataLoader(\n valid_ds, \n batch_size=CFG['valid_bs'],\n num_workers=CFG['num_workers'],\n shuffle=False,\n pin_memory=False,\n )\n \n tst_loader = torch.utils.data.DataLoader(\n test_ds, \n batch_size=CFG['valid_bs'],\n num_workers=CFG['num_workers'],\n shuffle=False,\n pin_memory=False,\n )\n\n device = torch.device(CFG['device'])\n model = CassvaImgClassifier(CFG['model_arch'], train.label.nunique()).to(device)\n \n val_preds = []\n tst_preds = []\n \n #for epoch in range(CFG['epochs']-3):\n for i, epoch in enumerate(CFG['used_epochs']): \n model.load_state_dict(torch.load('../input/project-cassava-leaf-disease/{}_fold_{}_{}'.format(CFG['model_arch'], fold, epoch)))\n \n with torch.no_grad():\n for _ in range(CFG['tta']):\n val_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, val_loader, device)]\n tst_preds += [CFG['weights'][i]/sum(CFG['weights'])/CFG['tta']*inference_one_epoch(model, tst_loader, device)]\n\n val_preds = np.mean(val_preds, axis=0) \n tst_preds = np.mean(tst_preds, axis=0) \n \n print('fold {} validation loss = {:.5f}'.format(fold, log_loss(valid_.label.values, val_preds)))\n print('fold {} validation accuracy = {:.5f}'.format(fold, (valid_.label.values==np.argmax(val_preds, axis=1)).mean()))\n \n del model\n torch.cuda.empty_cache()\n test['label'] = np.argmax(tst_preds, axis=1)\n test.head()\n test.to_csv('submission.csv', index=False)\n print(test)", "1198162": "This might help: https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215962", "1198266": "Thanks for your reply but can you have a look at my code and try to figure out what should I do any suggestions will be helpful.", "1198270": "Hard to debug without knowing what inference_one_epoch is. If you look at the recommendation in the post, you can try scoring the train dataset using your logic and see if you get a submission.csv. More than likely, there is an error in the code and it doesnt produce that file.", "1198769": "def inference_one_epoch(model, data_loader, device):\n model.eval()\n\n image_preds_all = []\n \n pbar = tqdm(enumerate(data_loader), total=len(data_loader))\n for step, (imgs) in pbar:\n imgs = imgs.to(device).float()\n \n image_preds = model(imgs) #output = model(input)\n image_preds_all += [torch.softmax(image_preds, 1).detach().cpu().numpy()]\n \n \n image_preds_all = np.concatenate(image_preds_all, axis=0)\n return image_preds_all\nthis is inference_one_epoch", "1199363": "did you try the recommendation?", "1200330": "Yes I changed the test path with train path and it produces a 21000 pedictions." }, "source": "meta" }