{"cells":[{"metadata":{},"cell_type":"markdown","source":"# Data preparation","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nimport json","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"%%time\n# prints cell time\n\n# Train data\nwith open('../input/herbarium-2020-fgvc7/nybg2020/train/metadata.json', \"r\", encoding=\"ISO-8859-1\") as file:\n train = json.load(file)\n\n# remove image_id, license and region_id columns because they are unnecessary\ntrain_img = pd.DataFrame(train['images']).drop(columns='license')\ntrain_ann = pd.DataFrame(train['annotations']).drop(columns=['image_id', 'region_id'])\n# final data frame\ntrain_df = train_img.merge(train_ann, on='id')\ntrain_df.head()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_df['category_id'].value_counts()","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"Target of Train set is `category_id` with **32093** different values","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"CATEGORY_CLASSES = 32093\n\n# set all labels by category_id\nfrom sklearn import preprocessing\n\nle = preprocessing.LabelEncoder()\nle.fit(train_df['category_id'])\ntrain_df['category_id_le'] = le.transform(train_df['category_id'])\nclass_map = dict(sorted(train_df[['category_id_le', 'category_id']].values.tolist()))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"%%time\n\n# Test data\nwith open('../input/herbarium-2020-fgvc7/nybg2020/test/metadata.json', \"r\", encoding=\"ISO-8859-1\") as file:\n test = json.load(file)\n\ntest_df = pd.DataFrame(test['images']).drop(columns='license')\ntest_df.head()","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Processing","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"import os\nimport random\nimport gc\ngc.enable()\nimport time\n\n# image loading\nimport cv2\nfrom PIL import Image\n\n# f1 score\nfrom sklearn.metrics import f1_score\n\n# progress bar\nfrom tqdm import tqdm\n\n# main library with neraul network for our image processing\nimport torch\nimport torch.nn as nn\nfrom torch.utils.data import DataLoader, Dataset\nimport torchvision.models as models\n\n# image transformations\nfrom albumentations import Compose, Normalize, Resize\nfrom albumentations.pytorch import ToTensorV2\n\n# set PyTorch processing with gpu if available\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# set seed for same result\nSEED = 999\n\nrandom.seed(SEED)\nos.environ['PYTHONHASHSEED'] = str(SEED)\nnp.random.seed(SEED)\ntorch.manual_seed(SEED)\ntorch.cuda.manual_seed(SEED)\n# necessary option for same behaviour\ntorch.backends.cudnn.deterministic = True","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"For any image size larger than *128x128* get **Memory error** shown below, so trained only for 128x128. I don’t know whether computing on my own computer will change the situation, because I have the same amount of RAM and even less GPU memory size.","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"# image transformation function\nHEIGHT = 200\nWIDTH = 200\n\ndef get_transforms():\n \n # Compose - use multiple transformations\n return Compose([\n Resize(HEIGHT, WIDTH),\n Normalize(\n mean=[0.485, 0.456, 0.406],\n std=[0.229, 0.224, 0.225],\n ),\n ToTensorV2(),\n ])\n","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# Train","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"Train set was already computed","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"# Train set class\nclass TrainDataset(Dataset):\n # Train data frame, category labels, transform function\n def __init__(self, df, labels, transform=None):\n self.df = df\n self.labels = labels\n self.transform = transform\n \n def __len__(self):\n return len(self.df)\n\n def __getitem__(self, idx):\n # load image and convert it to neccessary for 'albumentations' format\n file_name = self.df['file_name'].values[idx]\n file_path = f'../input/herbarium-2020-fgvc7/nybg2020/train/{file_name}'\n image = cv2.imread(file_path)\n image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)\n \n label = self.labels.values[idx]\n \n # transform and normalize image\n if self.transform:\n augmented = self.transform(image=image)\n image = augmented['image']\n \n return image, label","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# create 0 and 1 folds to then split for Train and Validation sets\nfrom sklearn.model_selection import StratifiedKFold\n\n# folds = train_df.sample(n=200000, random_state=0).reset_index(drop=True).copy()\nfolds = train_df.copy()\ntrain_labels = folds['category_id'].values\nkf = StratifiedKFold(n_splits=2)\nfor fold, (train_index, val_index) in enumerate(kf.split(folds.values, train_labels)):\n folds.loc[val_index, 'fold'] = int(fold)\nfolds['fold'] = folds['fold'].astype(int)\nfolds.to_csv('folds.csv', index=None)\nfolds.head()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"folds.shape","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"FOLD = 0\n# Train and Validation indices\ntrn_idx = folds[folds['fold'] != FOLD].index\nval_idx = folds[folds['fold'] == FOLD].index\n\n# Train and Validation data sets\ntrain_dataset = TrainDataset(folds.loc[trn_idx].reset_index(drop=True), \n folds.loc[trn_idx]['category_id'], \n transform=get_transforms())\nvalid_dataset = TrainDataset(folds.loc[val_idx].reset_index(drop=True), \n folds.loc[val_idx]['category_id'], \n transform=get_transforms())","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"batch_size = 512\n\n# decorators for data sets for easy iteration (must implement __len__ and __getitem__)\ntrain_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)\nvalid_loader = DataLoader(valid_dataset, batch_size=batch_size, shuffle=False)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# image recognition with Deep Residual Learning\nmodel = models.resnet18(pretrained=True)\n# max pooling\nmodel.avgpool = nn.AdaptiveAvgPool2d(1)\n# 2 layers classifier\nmodel.fc = nn.Linear(model.fc.in_features, CATEGORY_CLASSES)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# Train proccess\nfrom torch.optim import Adam\nfrom torch.optim.lr_scheduler import ReduceLROnPlateau\n\n# number of epochs\nn_epochs = 1\n# learning rate\nlr = 4e-4\n\nmodel.to(device)\n\n# stochastic optimization\noptimizer = Adam(model.parameters(), lr=lr, amsgrad=False)\n# optimize learning rate\nscheduler = ReduceLROnPlateau(optimizer, 'min', factor=0.75, patience=5, verbose=True, eps=1e-6)\n\n# cross entroy loss criterion\ncriterion = nn.CrossEntropyLoss()\n\nbest_score = 0.\nbest_loss = np.inf\n\nfor epoch in range(n_epochs):\n\n start_time = time.time()\n\n model.train()\n avg_loss = 0.\n \n # set gradients of model parameters to zero\n optimizer.zero_grad()\n \n # training\n for i, (images, labels) in tqdm(enumerate(train_loader)):\n\n # load to device\n images = images.to(device)\n labels = labels.to(device)\n \n # compute output\n y_preds = model(images)\n loss = criterion(y_preds, labels)\n \n # backward propogation\n loss.backward()\n # adjust weights\n optimizer.step()\n optimizer.zero_grad()\n\n avg_loss += loss.item() / len(train_loader)\n \n # enable prediction mode\n model.eval()\n avg_val_loss = 0.\n preds = np.zeros((len(valid_dataset)))\n \n # validating\n for i, (images, labels) in enumerate(valid_loader):\n\n images = images.to(device)\n labels = labels.to(device)\n \n with torch.no_grad():\n y_preds = model(images)\n\n preds[i * batch_size: (i+1) * batch_size] = y_preds.argmax(1).to('cpu').numpy()\n\n loss = criterion(y_preds, labels)\n avg_val_loss += loss.item() / len(valid_loader)\n \n # optimize learning rate\n scheduler.step(avg_val_loss)\n \n # epoch score\n score = f1_score(folds.loc[val_idx]['category_id'].values, preds, average='macro')\n\n elapsed = time.time() - start_time\n\n print(f'Epoch {epoch+1} avg_train_loss: {avg_loss:.4f} avg_val_loss: {avg_val_loss:.4f} F1: {score:.6f} time: {elapsed:.0f}s')\n\n if score>best_score:\n best_score = score\n print(f'Epoch {epoch+1} save best score: {best_score:.6f} Model')\n torch.save(model.state_dict(), f'fold_{FOLD}_best_score.pth')\n\n if avg_val_loss