{"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":"# Import Libraries","metadata":{}},{"cell_type":"code","source":"import pandas as pd\nimport numpy as np\nimport glob\nimport os\nimport matplotlib.pyplot as plt\nimport zipfile\nimport torch.nn as nn\nimport torch\nfrom torch import optim\nimport torch.nn.functional as F\nfrom torch.utils.data import Dataset,DataLoader\nfrom torchvision import transforms\nfrom torchvision import models\nfrom PIL import Image\nfrom collections import OrderedDict\nfrom tqdm import tqdm\nfrom sklearn import model_selection\nfrom statistics import mode","metadata":{"execution":{"iopub.status.busy":"2022-11-12T06:01:13.061211Z","iopub.execute_input":"2022-11-12T06:01:13.061835Z","iopub.status.idle":"2022-11-12T06:01:13.072344Z","shell.execute_reply.started":"2022-11-12T06:01:13.061794Z","shell.execute_reply":"2022-11-12T06:01:13.071144Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Define All Variables","metadata":{}},{"cell_type":"code","source":"base_dir = '../input/dogs-vs-cats'\ntrain_zip = os.path.join(base_dir,'train.zip')\ntest_zip = os.path.join(base_dir,'test1.zip')\n\n# Device\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n\n# Model\nmodel = models.resnet18(pretrained=True)\nepochs = 20\nlearning_rate = 0.1\nbatch_size = 64\ncross_val_splits=5\nimg_size=128","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:37:41.616679Z","iopub.execute_input":"2022-11-12T05:37:41.617139Z","iopub.status.idle":"2022-11-12T05:37:44.492352Z","shell.execute_reply.started":"2022-11-12T05:37:41.617109Z","shell.execute_reply":"2022-11-12T05:37:44.491349Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Extracting ZIP Files","metadata":{}},{"cell_type":"code","source":"with zipfile.ZipFile(train_zip,\"r\") as z:\n    z.extractall()\n\nwith zipfile.ZipFile(test_zip,\"r\") as z:\n    z.extractall()","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:37:44.494132Z","iopub.execute_input":"2022-11-12T05:37:44.494663Z","iopub.status.idle":"2022-11-12T05:38:02.721614Z","shell.execute_reply.started":"2022-11-12T05:37:44.494625Z","shell.execute_reply":"2022-11-12T05:38:02.718861Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Checking Sample DataFrame","metadata":{}},{"cell_type":"code","source":"sample_df = pd.read_csv(os.path.join(base_dir,'sampleSubmission.csv'))\nsample_df.head()","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:02.726343Z","iopub.execute_input":"2022-11-12T05:38:02.727168Z","iopub.status.idle":"2022-11-12T05:38:02.795564Z","shell.execute_reply.started":"2022-11-12T05:38:02.727120Z","shell.execute_reply":"2022-11-12T05:38:02.792407Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Creating Dataframes for Training","metadata":{}},{"cell_type":"code","source":"train_dir = './train'\ntest_dir = './test1'\n\ntrain_df = pd.DataFrame()\ntest_df = pd.DataFrame()\ntrain_df['file_path'] = glob.glob('./train/*.*')\ntrain_df['file_name'] = [os.path.basename(val) for val in train_df['file_path']]\ntrain_df['file_id'] = train_df['file_name'].str.split('.').str[1]\ntrain_df['label'] = train_df['file_name'].str.split('.').str[0]\ntrain_df['label'] = train_df['label'].map({'cat':0,'dog':1})\n\n\ntest_df['file_path'] = glob.glob('./test1/*.*')\ntest_df['file_name'] = [os.path.basename(val) for val in test_df['file_path']]\ntest_df['file_id'] = test_df['file_name'].str.split('.').str[0]\ntest_df['label'] = 9999","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:02.797452Z","iopub.execute_input":"2022-11-12T05:38:02.798444Z","iopub.status.idle":"2022-11-12T05:38:03.113917Z","shell.execute_reply.started":"2022-11-12T05:38:02.798402Z","shell.execute_reply":"2022-11-12T05:38:03.112895Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Creating Folds for train data","metadata":{}},{"cell_type":"code","source":"train_df['kfold'] = -1\ny = train_df.label.values\n\nkf = model_selection.StratifiedKFold(n_splits=cross_val_splits, shuffle=True, random_state=42)\n\nfor f,(t_idx,val_idx) in enumerate(kf.split(X=train_df, y=y)):\n    train_df.loc[val_idx, 'kfold'] = f\n    \ntrain_df.to_csv(\"train_folds.csv\", index=False)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:03.115243Z","iopub.execute_input":"2022-11-12T05:38:03.115586Z","iopub.status.idle":"2022-11-12T05:38:03.178472Z","shell.execute_reply.started":"2022-11-12T05:38:03.115553Z","shell.execute_reply":"2022-11-12T05:38:03.177580Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Loading Train dataframe with kfold index","metadata":{}},{"cell_type":"code","source":"train_df = pd.read_csv('./train_folds.csv')\ntrain_df.head()","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:03.179718Z","iopub.execute_input":"2022-11-12T05:38:03.180696Z","iopub.status.idle":"2022-11-12T05:38:03.217930Z","shell.execute_reply.started":"2022-11-12T05:38:03.180657Z","shell.execute_reply":"2022-11-12T05:38:03.217006Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Define DataSet Class","metadata":{}},{"cell_type":"code","source":"class CatDogDataset(Dataset):\n    def __init__(self,df , mode='train',transform = None):\n        self.df = df\n        self.mode= mode\n        self.transform = transform\n    \n    def __len__(self):\n        return len(self.df)\n    \n    def __getitem__(self,idx):\n        img = Image.open(self.df.loc[:,'file_path'][idx])\n        label = self.df.loc[:,'label'][idx]\n        \n        if self.transform:\n            img = self.transform(img)\n            \n        if self.mode=='train':\n            img = img.numpy()\n            return img.astype('float32'), label\n        else:\n            img = img.numpy()\n            return img.astype('float32') , label","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:03.219580Z","iopub.execute_input":"2022-11-12T05:38:03.219967Z","iopub.status.idle":"2022-11-12T05:38:03.227677Z","shell.execute_reply.started":"2022-11-12T05:38:03.219932Z","shell.execute_reply":"2022-11-12T05:38:03.226446Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Define Transformations","metadata":{}},{"cell_type":"code","source":"train_transform = transforms.Compose([transforms.Resize((img_size,img_size)),\n                                     transforms.ToTensor(),\n                                     transforms.Normalize([0.485,0.456,0.406],[0.229,0.224,0.225]),\n                                     transforms.RandomHorizontalFlip(0.5)])\n\nval_transform = transforms.Compose([transforms.Resize((img_size,img_size)),\n                                     transforms.ToTensor(),\n                                     transforms.Normalize([0.485,0.456,0.406],[0.229,0.224,0.225])])\n\ntest_transform = transforms.Compose([transforms.Resize((img_size,img_size)),\n                                     transforms.ToTensor(),\n                                     transforms.Normalize([0.485,0.456,0.406],[0.229,0.224,0.225])])","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:03.229330Z","iopub.execute_input":"2022-11-12T05:38:03.229674Z","iopub.status.idle":"2022-11-12T05:38:03.243593Z","shell.execute_reply.started":"2022-11-12T05:38:03.229640Z","shell.execute_reply":"2022-11-12T05:38:03.242545Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Creating Sample Loader for visualization","metadata":{}},{"cell_type":"code","source":"sample_set = CatDogDataset(train_df,mode = 'train' ,transform = train_transform)\nsample_loader = DataLoader(sample_set, batch_size = batch_size, shuffle=True, num_workers=2)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:03.248055Z","iopub.execute_input":"2022-11-12T05:38:03.248932Z","iopub.status.idle":"2022-11-12T05:38:03.254931Z","shell.execute_reply.started":"2022-11-12T05:38:03.248898Z","shell.execute_reply":"2022-11-12T05:38:03.253893Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Checking Shape\nX,y = next(iter(sample_loader))\nX.shape,sample_loader.batch_size","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:03.257989Z","iopub.execute_input":"2022-11-12T05:38:03.258244Z","iopub.status.idle":"2022-11-12T05:38:04.237155Z","shell.execute_reply.started":"2022-11-12T05:38:03.258214Z","shell.execute_reply":"2022-11-12T05:38:04.236003Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# View a batch of data","metadata":{}},{"cell_type":"code","source":"def display_batch():\n    figure = plt.figure(figsize=(15, 15))\n    cols, rows = 8, 8\n    for i in range(1, cols * rows):\n        img,label = next(iter(sample_loader))\n        figure.add_subplot(rows, cols, i)\n        plt.title(label[i])\n        plt.axis(\"off\")\n        plt.imshow(img[i].permute(1, 2, 0),cmap=\"gray\")\n    plt.show()\n\n# display_batch()","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:04.238965Z","iopub.execute_input":"2022-11-12T05:38:04.240232Z","iopub.status.idle":"2022-11-12T05:38:04.248235Z","shell.execute_reply.started":"2022-11-12T05:38:04.240187Z","shell.execute_reply":"2022-11-12T05:38:04.247058Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Alter the last layer for training","metadata":{}},{"cell_type":"code","source":"for param in model.parameters():\n    param.requires_grad = False \nfc = nn.Sequential(OrderedDict([\n    ('fc1', nn.Linear(512,100)),\n    ('relu', nn.ReLU()),\n    ('fc2', nn.Linear(100,2)),\n    ('output', nn.LogSoftmax(dim=1))\n]))\n\nmodel.fc = fc\nmodel = model.to(device)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:04.249814Z","iopub.execute_input":"2022-11-12T05:38:04.250297Z","iopub.status.idle":"2022-11-12T05:38:08.025161Z","shell.execute_reply.started":"2022-11-12T05:38:04.250260Z","shell.execute_reply":"2022-11-12T05:38:08.024143Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Defina Optimizer & Loss","metadata":{}},{"cell_type":"code","source":"optimizer = optim.Adam(model.fc.parameters(), lr=learning_rate)\ncriterion = nn.NLLLoss() \nscheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer,patience=5,verbose=True)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:08.026659Z","iopub.execute_input":"2022-11-12T05:38:08.027258Z","iopub.status.idle":"2022-11-12T05:38:08.033745Z","shell.execute_reply.started":"2022-11-12T05:38:08.027220Z","shell.execute_reply":"2022-11-12T05:38:08.032262Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Define Training Loop","metadata":{}},{"cell_type":"code","source":"def train(dataset,dataloader,criterion,model,optimizer):\n    model.train()\n    for bi,(img,lab) in tqdm(enumerate(dataloader),total = int(len(dataset)/dataloader.batch_size)):\n        img,lab = img.to(device),lab.to(device)\n        optimizer.zero_grad()\n        outputs = model(img)\n        loss = criterion(outputs,lab)\n        loss.backward()\n        optimizer.step()","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:08.035496Z","iopub.execute_input":"2022-11-12T05:38:08.035851Z","iopub.status.idle":"2022-11-12T05:38:08.046227Z","shell.execute_reply.started":"2022-11-12T05:38:08.035798Z","shell.execute_reply":"2022-11-12T05:38:08.045388Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Defina Evaluate Function","metadata":{}},{"cell_type":"code","source":"def evaluate(dataset,dataloader,criterion,model):\n    model.eval()\n    final_loss = 0\n    counter = 0\n    for bi,(img,lab) in tqdm(enumerate(dataloader),total = int(len(dataset)/dataloader.batch_size)):\n        counter = counter+1\n        img,lab = img.to(device),lab.to(device)\n        outputs = model(img)\n        loss = criterion(outputs,lab)\n        final_loss+=loss\n    return final_loss/counter\n        ","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:08.047830Z","iopub.execute_input":"2022-11-12T05:38:08.048231Z","iopub.status.idle":"2022-11-12T05:38:08.058576Z","shell.execute_reply.started":"2022-11-12T05:38:08.048198Z","shell.execute_reply":"2022-11-12T05:38:08.057580Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Running Training & Evaluation","metadata":{}},{"cell_type":"code","source":"def train_fold(fold):\n    print(f\"For Fold={fold}\")\n    for epoch in range(epochs):\n        print(f\"Epoch={epoch}\")\n        train_dataframe = train_df.query('kfold!=@fold').reset_index(drop=True)\n        val_dataframe = train_df.query('kfold==@fold').reset_index(drop=True)\n\n        trainset = CatDogDataset(train_dataframe,mode = 'train' ,transform = train_transform)\n        train_loader = DataLoader(trainset, batch_size = batch_size, shuffle=True, num_workers=0)\n\n        valset = CatDogDataset(val_dataframe,mode = 'test' ,transform = val_transform)\n        val_loader = DataLoader(valset,batch_size = batch_size, shuffle=False, num_workers=0)\n\n\n        train(trainset,train_loader,criterion,model,optimizer)\n        val_score = evaluate(valset,val_loader,criterion,model)\n        scheduler.step(val_score)\n        torch.save(model.state_dict(),f\"resnet18_fold_{fold}.bin\")","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:08.059924Z","iopub.execute_input":"2022-11-12T05:38:08.060323Z","iopub.status.idle":"2022-11-12T05:38:08.070640Z","shell.execute_reply.started":"2022-11-12T05:38:08.060290Z","shell.execute_reply":"2022-11-12T05:38:08.069666Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_fold(0)\ntrain_fold(1)\ntrain_fold(2)\ntrain_fold(3)\ntrain_fold(4)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:38:08.071861Z","iopub.execute_input":"2022-11-12T05:38:08.072342Z","iopub.status.idle":"2022-11-12T05:49:13.361469Z","shell.execute_reply.started":"2022-11-12T05:38:08.072309Z","shell.execute_reply":"2022-11-12T05:49:13.360330Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Define Prediction Function","metadata":{}},{"cell_type":"code","source":"def prediction_fold(fold):\n    model.load_state_dict(torch.load(f'./resnet18_fold_{fold}.bin'))\n    model.eval()\n    pred_list = []\n    test_dataset = CatDogDataset(test_df,mode = 'test' ,transform = test_transform)\n    testloader = DataLoader(test_dataset, batch_size = batch_size, shuffle=False, num_workers=0)\n    for x, fn in tqdm(testloader):\n        with torch.no_grad():\n            x = x.to(device)\n            output = model(x)\n            pred = torch.argmax(output, dim=1)\n            pred_list += [p.item() for p in pred]\n    return pred_list\n    ","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:49:13.363158Z","iopub.execute_input":"2022-11-12T05:49:13.363806Z","iopub.status.idle":"2022-11-12T05:49:13.371428Z","shell.execute_reply.started":"2022-11-12T05:49:13.363768Z","shell.execute_reply":"2022-11-12T05:49:13.370260Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"p1 = prediction_fold(0)\np2 = prediction_fold(1)\np3 = prediction_fold(2)\np4 = prediction_fold(3)\np5 = prediction_fold(4)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T05:49:13.372923Z","iopub.execute_input":"2022-11-12T05:49:13.373361Z","iopub.status.idle":"2022-11-12T05:52:11.054189Z","shell.execute_reply.started":"2022-11-12T05:49:13.373327Z","shell.execute_reply":"2022-11-12T05:52:11.053195Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"final_predictions = [mode(val) for val in list(zip(p1,p2,p3,p4,p5))]","metadata":{"execution":{"iopub.status.busy":"2022-11-12T06:02:43.241852Z","iopub.execute_input":"2022-11-12T06:02:43.242492Z","iopub.status.idle":"2022-11-12T06:02:43.305698Z","shell.execute_reply.started":"2022-11-12T06:02:43.242444Z","shell.execute_reply":"2022-11-12T06:02:43.304786Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission = sample_df[['id']].merge(pd.DataFrame({'id':test_df['file_id'].astype(int),'label':final_predictions}),how='left',on='id')\nsubmission.to_csv('preds_resnet18.csv', index=False)","metadata":{"execution":{"iopub.status.busy":"2022-11-12T06:02:46.838295Z","iopub.execute_input":"2022-11-12T06:02:46.838847Z","iopub.status.idle":"2022-11-12T06:02:46.878908Z","shell.execute_reply.started":"2022-11-12T06:02:46.838785Z","shell.execute_reply":"2022-11-12T06:02:46.877862Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}