{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        print(os.path.join(dirname, filename))\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"GPUを起動させたら、全てのコードを実行し直さないといけない？\n\nStratifiedKFoldで分けたこと  \nTrainDatasetがやっていること  \nDataLoaderがやっていること  \n"},{"metadata":{"trusted":true},"cell_type":"code","source":"!pip install ttach\n!pip install timm","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"cell_type":"code","source":"#import geffnet\nimport sys\nimport gc\nimport os\nimport random\nimport time\nfrom contextlib import contextmanager\nfrom collections import defaultdict, Counter\nfrom  torch.cuda.amp import autocast, GradScaler \nimport cv2\nfrom PIL import Image\nimport numpy as np\nimport pandas as pd\nimport scipy as sp\nimport sklearn.metrics\nfrom sklearn.metrics import accuracy_score\nfrom sklearn.model_selection import StratifiedKFold\nfrom sklearn.metrics import log_loss\nfrom functools import partial\nfrom tqdm import tqdm\nimport ttach as tta##TTA\nimport timm\nimport matplotlib as mpl\nmpl.use ( 'Agg') # must be written both in import intermediate\nimport matplotlib.pyplot as plt\nimport sklearn.metrics as metric\nimport torch\nimport torch.nn as nn\nfrom torch.optim import Adam, SGD\nfrom torch.optim.lr_scheduler import CosineAnnealingLR, ReduceLROnPlateau\nfrom torch.utils.data import DataLoader, Dataset\nimport torchvision.models as models\nfrom albumentations import Compose, Normalize, HorizontalFlip, VerticalFlip,RandomGamma, RandomRotate90,GaussNoise,Cutout\nfrom albumentations.pytorch import ToTensor","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"import datetime\ndt_now = datetime.datetime.now()\ndt_now_ = str(dt_now).replace(\" \",\"_\")\nprint(\"実験開始\",dt_now_)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"import pandas as pd \ndf = pd.read_csv(\"../input/cassava-leaf-disease-classification/train.csv\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"df","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# 訓練データの画像の確認\npath = \"../input/cassava-leaf-disease-classification/train_images/100042118.jpg\"\nimport cv2\nimport matplotlib.pyplot as plt\n%matplotlib inline\n\nimg = cv2.imread(path)\nim_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)\n\nprint(img.shape)\n\n#matplotlibではカラーマップを与えないといけない。よく使うcmapはgrayとjet\nplt.imshow(im_gray,cmap=\"gray\")\nplt.show()\nplt.imshow(im_gray,cmap=\"jet\")\nplt.show()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class CFG:\n    debug=True\n    n_classes = 5\n    lr=1e-4\n    batch_size=8\n    epochs=1# you can train more epochs\n    seed=777\n    n_fold=4\n    warmup=-1\n    device=0\n    amp = True\n    amp_inf = False\n    smooth = False\n    smooth_alpha = 0.1\n    efnet_num = 10##:0,1,2\n    drop_rate = 0.25\n    crop = False##bool\n    psuedo_label = False\n    pseudo_predict = \"2020-11-06_14:50:43.385109_predict.csv\"\n    TTA=False\n    Attention=False\n    white = False#ごましおになる\n    model_name = \"tf_efficientnet_b0_ns\"\n    zoom = True\nSIZE = 512\n\nmodel_names = [\"tf_efficientnet_b0_ns\",\"vit_base_resnet26d_224\",\"resnest50d\",\"vit_base_patch32_384\"]","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"#乱数を固定する\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\nseed_torch(seed=42)\ntorch.cuda.set_device(CFG.device)\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# train, testデータに分ける\nfrom sklearn.model_selection import train_test_split\ntrain, test = train_test_split(df, test_size=0.9,stratify = df[\"label\"], random_state=2020) #stratifyは層化\ntrain = train.reset_index(drop=True) #reset_indexで歯抜けになったindexを振り直している\ntest = test.reset_index(drop=True)\nprint(train[\"label\"].value_counts())\nprint(test[\"label\"].value_counts())","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class efnet_model(nn.Module):\n\n    def __init__(self):\n        super().__init__()\n        if CFG.efnet_num ==0:\n            self.model = geffnet.efficientnet_b0(pretrained=True, drop_rate=CFG.drop_rate)\n        elif CFG.efnet_num ==1:\n            self.model = geffnet.efficientnet_b1(pretrained=True, drop_rate=CFG.drop_rate)\n        elif CFG.efnet_num ==2:\n            self.model = geffnet.efficientnet_b2(pretrained=True, drop_rate=CFG.drop_rate)\n        elif CFG.efnet_num ==10:\n            self.model = timm.create_model(CFG.model_name, pretrained=True, num_classes=CFG.n_classes)\n            #self.head = nn.Sequential(nn.AdaptiveAvgPool2d(1),nn.Flatten(),nn.Linear(nc,CFG.n_classes))\n        elif CFG.efnet_num<3:\n            self.model.classifier = nn.Linear(self.model.classifier.in_features, 2)\n\n    def forward(self, x):\n        \n        if CFG.efnet_num ==10:\n            x = self.model(x)\n        else:\n            x = self.model(x)#ベースのモデルの流れに同じ\n        return x","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"if CFG.debug:\n    folds = train.sample(n=200,random_state=CFG.seed).reset_index(drop=True).copy()\nelse:\n    folds = train.copy()\ntrain_labels = folds[\"label\"].values\nkf = StratifiedKFold(n_splits=CFG.n_fold, shuffle=True, random_state=CFG.seed)\nfor fold, (train_index, val_index) in enumerate(kf.split(folds.values, train_labels)):\n    print(\"num_train,val\",len(train_index),len(val_index),len(val_index)+len(train_index))\n    folds.loc[val_index, 'fold'] = int(fold)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"folds","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class TrainDataset(Dataset):\n    def __init__(self, df,transform1=None, transform2=None):\n        self.df = df\n        self.transform = transform1\n        self.transform_ = transform2\n        \n    def __len__(self):\n        return len(self.df)\n\n    def __getitem__(self, idx):\n        #file_path = self.df['file'].values[idx]\n        path = self.df['image_id'].values[idx]\n        file_path = \"/kaggle/input/cassava-leaf-disease-classification/train_images/{}\".format(path)\n        image = cv2.imread(file_path)\n        if CFG.crop:\n            image = crop_object(image)\n        try:\n            image = cv2.resize(image,(SIZE,SIZE))\n        except Exception as e:\n            print(file_path)\n        label_ = self.df[\"label\"].values[idx]\n        if self.transform:\n            image = self.transform(image=image)['image']\n        if self.transform_:\n            image = self.transform_(image=image)['image']\n        \n        label = torch.tensor(label_)\n        #print(label_,type(label_),label,label.size())\n        \n        return image, label","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def get_transforms1(*, data):\n    if data == 'train':\n        return Compose([\n            HorizontalFlip(p=0.5),\n            VerticalFlip(p=0.5),\n            #GaussNoise(p=0.5),\n            #RandomRotate90(p=0.5),\n            #RandomGamma(p=0.5),\n            #RandomAugMix(severity=3, width=3, alpha=1., p=0.5),\n            #GaussianBlur(p=0.5),\n            #GridMask(num_grid=3, p=0.3),\n            #Cutout(p=0.5),\n            Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225])\n        ])\n    elif data == 'valid':\n        return Compose([\n            Normalize(\n                mean=[0.485, 0.456, 0.406],\n                std=[0.229, 0.224, 0.225],\n            )\n        ])\ndef to_tensor(*args):\n        return Compose([\n            ToTensor()\n        ])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"#学習のループ\ndef train_fn(fold):\n    print(f\"### fold: {fold} ###\")\n    trn_idx = folds[folds['fold'] != fold].index\n    val_idx = folds[folds['fold'] == fold].index\n    train_df = folds.loc[trn_idx].reset_index(drop=True)\n    valid_df = folds.loc[val_idx].reset_index(drop=True)\n    train_dataset = TrainDataset(folds.loc[trn_idx].reset_index(drop=True), \n                                 transform1=get_transforms1(data='train'),transform2=to_tensor())#\n    valid_dataset = TrainDataset(folds.loc[val_idx].reset_index(drop=True), \n                                 transform1=get_transforms1(data='valid'),transform2=to_tensor())#\n    train_loader = DataLoader(train_dataset, batch_size=CFG.batch_size, shuffle=True, num_workers=4)\n    valid_loader = DataLoader(valid_dataset, batch_size=CFG.batch_size, shuffle=False, num_workers=4)\n    model = efnet_model()\n    model.to(device)\n    optimizer = Adam(model.parameters(), lr=CFG.lr, amsgrad=False)\n    criterion = nn.CrossEntropyLoss()#weight = class_weight\n    softmax = nn.Softmax(dim = 1)\n    for epoch in range(CFG.epochs):\n        start_time = time.time()\n        model.train()\n        avg_loss = 0.\n        tk0 = tqdm(enumerate(train_loader), total=len(train_loader))\n        for i, (images, labels) in tk0:\n          images = images.to(device)\n          labels = labels.to(device)\n          optimizer.zero_grad()\n          y_preds = model(images.float())\n          loss = criterion(y_preds, labels.long())\n          loss.backward()\n          optimizer.step()\n          avg_loss += loss.item() / len(train_loader)\n        model.eval()\n        avg_val_loss = 0.\n        valid_labels = []\n        preds = []\n        tk1 = tqdm(enumerate(valid_loader), total=len(valid_loader))\n        for i, (images, labels) in tk1:\n          images = images.to(device)\n          labels = labels.to(device)\n          with torch.no_grad():\n             y_preds = model(images.float())\n             loss = criterion(y_preds,labels.long())\n          valid_labels.append(labels.to('cpu').detach().numpy().copy())\n          y_preds = softmax(y_preds)\n          preds.append(y_preds.to('cpu').detach().numpy().copy())\n          avg_val_loss += loss.item() / len(valid_loader)\n        preds = np.concatenate(preds)\n        valid_labels = np.concatenate(valid_labels)\n        torch.save(model.state_dict(), f'fold{fold}_{dt_now_}_baseline.pth')\n        return preds,valid_labels\n","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def auc(predict,labels):\n    pass","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"predict = []\nlabels = []\nfor fold in range(CFG.n_fold):\n  _pred,_label = train_fn(fold)\n  predict.append(_pred)\n  labels.append(_label)\npredict = np.concatenate(predict)#予測\nlabels = np.concatenate(labels)#正解ラベル\nscore = auc(predict,labels)#aucはラベルに置き換えなくても使える指標\n\nprint(predict)\nprint(labels)\nprint(len(labels)) #4つのfoldの50の予測を合体させたのがpredict200個であり、labelsは正解ラベル\nprint(score)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"","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}