{"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"},"kaggle":{"accelerator":"none","dataSources":[{"sourceType":"competition","sourceId":3004,"databundleVersionId":861823}],"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"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","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"/kaggle/input/competitions/digit-recognizer/sample_submission.csv\n/kaggle/input/competitions/digit-recognizer/train.csv\n/kaggle/input/competitions/digit-recognizer/test.csv","metadata":{}},{"cell_type":"code","source":"train = pd.read_csv('/kaggle/input/competitions/digit-recognizer/train.csv')\ntest = pd.read_csv('/kaggle/input/competitions/digit-recognizer/test.csv')\n","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"X = train.drop(columns=['label'])\ny = train['label']","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from torch.utils.data import Dataset, DataLoader\nimport os\nimport torch\n\nclass ImageDataset(Dataset):\n    def __init__(self, path, transform=None):\n        super().__init__()\n        self.transform = transform\n        self.file = pd.read_csv(path)\n        \n        self.targets = torch.tensor(self.file['label'].values, dtype=torch.long)\n        self.pixels = torch.tensor(self.file.drop(columns=['label']).values, dtype=torch.float32)\n        \n    def __getitem__(self, idx):\n        pixels = self.pixels[idx]\n        target = self.targets[idx]\n\n        pixels = pixels.view(1, 28, 28)\n\n        if self.transform:\n            pixels = self.transform(pixels)\n        \n        return pixels, target\n\n        return pixels, target\n    def __len__(self):\n        return len(self.targets)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import torchvision.transforms.v2 as tsf\n\ntransform = tsf.Compose([\n    tsf.RandomRotation(degrees=10),\n    tsf.RandomAffine(degrees=0, translate=(0.1, 0.1), scale=(0.9, 1.1))\n])","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_data = ImageDataset('/kaggle/input/competitions/digit-recognizer/train.csv', transform=transform)\ntrain_data = DataLoader(train_data, batch_size=32, shuffle=True)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import torch.nn as nn\n\nmodel = nn.Sequential(\n    nn.Conv2d(1, 32, kernel_size=3, padding=1, bias=False),\n    nn.BatchNorm2d(32),\n    nn.ReLU(),\n    nn.Conv2d(32, 32, kernel_size=3, padding=1, bias=False),\n    nn.BatchNorm2d(32),\n    nn.ReLU(),\n    nn.MaxPool2d(2),\n    nn.Dropout2d(0.2),\n\n    nn.Conv2d(32, 64, kernel_size=3, padding=1, bias=False),\n    nn.BatchNorm2d(64),\n    nn.ReLU(),\n    nn.Conv2d(64, 64, kernel_size=3, padding=1, bias=False),\n    nn.BatchNorm2d(64),\n    nn.ReLU(),\n    nn.MaxPool2d(2), \n    nn.Dropout2d(0.3),\n    \n    nn.Flatten(),\n    nn.Linear(64 * 7 * 7, 128),\n    nn.BatchNorm1d(128),\n    nn.ReLU(),\n    nn.Dropout(0.4),\n    nn.Linear(128, 10)\n)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import torch\nimport torch.nn as nn\nimport torch.optim as optim\nfrom tqdm import tqdm\n\nloss_func = nn.CrossEntropyLoss() \noptimizer = optim.AdamW(model.parameters(), lr=0.001, weight_decay=0.001)\n\nepochs = 30\nmodel.train()\n\nfor e in range(epochs):\n    running_loss = 0.0\n    train_tqdm = tqdm(train_data, leave=True)\n\n    for x_train, y_train in train_tqdm:\n        x_train = x_train.view(-1, 1, 28, 28)\n        \n        pred = model(x_train) \n        \n        loss = loss_func(pred, y_train)\n\n        optimizer.zero_grad()\n        loss.backward()\n        optimizer.step()\n\n        running_loss += loss.item()\n        train_tqdm.set_description(f'Epoch [{e+1}/{epochs}], loss={loss.item():.4f}')\n\n    print(f\"Epoch {e+1} finished. Avg Loss: {running_loss / len(train_data):.4f}\")\n\ntorch.save(model.state_dict(), 'model_123.pth')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"Epoch [1/30], loss=0.3364: 100%|██████████| 1313/1313 [01:33<00:00, 14.05it/s]\nEpoch 1 finished. Avg Loss: 0.1244\nEpoch [2/30], loss=0.3095: 100%|██████████| 1313/1313 [01:32<00:00, 14.17it/s]\nEpoch 2 finished. Avg Loss: 0.1073\nEpoch [3/30], loss=0.1043: 100%|██████████| 1313/1313 [01:32<00:00, 14.22it/s]\nEpoch 3 finished. Avg Loss: 0.0936\nEpoch [4/30], loss=0.0465: 100%|██████████| 1313/1313 [01:31<00:00, 14.34it/s]\nEpoch 4 finished. Avg Loss: 0.0871\nEpoch [5/30], loss=0.0137: 100%|██████████| 1313/1313 [01:31<00:00, 14.35it/s]\nEpoch 5 finished. Avg Loss: 0.0807\nEpoch [6/30], loss=0.2454: 100%|██████████| 1313/1313 [01:32<00:00, 14.22it/s]\nEpoch 6 finished. Avg Loss: 0.0748\nEpoch [7/30], loss=0.0078: 100%|██████████| 1313/1313 [01:33<00:00, 14.09it/s]\nEpoch 7 finished. Avg Loss: 0.0733\nEpoch [8/30], loss=0.0661: 100%|██████████| 1313/1313 [01:32<00:00, 14.18it/s]\nEpoch 8 finished. Avg Loss: 0.0705\nEpoch [9/30], loss=0.0106: 100%|██████████| 1313/1313 [01:32<00:00, 14.27it/s]\nEpoch 9 finished. Avg Loss: 0.0689\nEpoch [10/30], loss=0.1083: 100%|██████████| 1313/1313 [01:32<00:00, 14.21it/s]\nEpoch 10 finished. Avg Loss: 0.0651\nEpoch [11/30], loss=0.0003: 100%|██████████| 1313/1313 [01:31<00:00, 14.28it/s]\nEpoch 11 finished. Avg Loss: 0.0614\nEpoch [12/30], loss=0.0721: 100%|██████████| 1313/1313 [01:31<00:00, 14.31it/s]\nEpoch 12 finished. Avg Loss: 0.0622\nEpoch [13/30], loss=0.1418: 100%|██████████| 1313/1313 [01:32<00:00, 14.25it/s]\nEpoch 13 finished. Avg Loss: 0.0637\nEpoch [14/30], loss=0.0071: 100%|██████████| 1313/1313 [01:33<00:00, 14.11it/s]\nEpoch 14 finished. Avg Loss: 0.0591\nEpoch [15/30], loss=0.0205: 100%|██████████| 1313/1313 [01:32<00:00, 14.27it/s]\nEpoch 15 finished. Avg Loss: 0.0584\nEpoch [16/30], loss=0.0062: 100%|██████████| 1313/1313 [01:32<00:00, 14.17it/s]\nEpoch 16 finished. Avg Loss: 0.0535\nEpoch [17/30], loss=0.0021: 100%|██████████| 1313/1313 [01:32<00:00, 14.21it/s]\nEpoch 17 finished. Avg Loss: 0.0561\nEpoch [18/30], loss=0.0230: 100%|██████████| 1313/1313 [01:31<00:00, 14.30it/s]\nEpoch 18 finished. Avg Loss: 0.0517\nEpoch [19/30], loss=0.0542: 100%|██████████| 1313/1313 [01:31<00:00, 14.27it/s]\nEpoch 19 finished. Avg Loss: 0.0552\nEpoch [20/30], loss=0.1897: 100%|██████████| 1313/1313 [01:32<00:00, 14.24it/s]\nEpoch 20 finished. Avg Loss: 0.0511\nEpoch [21/30], loss=0.2345: 100%|██████████| 1313/1313 [01:31<00:00, 14.36it/s]\nEpoch 21 finished. Avg Loss: 0.0492\nEpoch [22/30], loss=0.0015: 100%|██████████| 1313/1313 [01:31<00:00, 14.35it/s]\nEpoch 22 finished. Avg Loss: 0.0490\nEpoch [23/30], loss=0.0042: 100%|██████████| 1313/1313 [01:32<00:00, 14.26it/s]\nEpoch 23 finished. Avg Loss: 0.0494\nEpoch [24/30], loss=0.1129: 100%|██████████| 1313/1313 [01:31<00:00, 14.30it/s]\nEpoch 24 finished. Avg Loss: 0.0468\nEpoch [25/30], loss=0.0007: 100%|██████████| 1313/1313 [01:32<00:00, 14.22it/s]\nEpoch 25 finished. Avg Loss: 0.0480\nEpoch [26/30], loss=0.0198: 100%|██████████| 1313/1313 [01:31<00:00, 14.41it/s]\nEpoch 26 finished. Avg Loss: 0.0482\nEpoch [27/30], loss=0.0055: 100%|██████████| 1313/1313 [01:32<00:00, 14.22it/s]\nEpoch 27 finished. Avg Loss: 0.0483\nEpoch [28/30], loss=0.1905: 100%|██████████| 1313/1313 [01:31<00:00, 14.30it/s]\nEpoch 28 finished. Avg Loss: 0.0467\nEpoch [29/30], loss=0.0174: 100%|██████████| 1313/1313 [01:30<00:00, 14.44it/s]\nEpoch 29 finished. Avg Loss: 0.0453\nEpoch [30/30], loss=0.0218: 100%|██████████| 1313/1313 [01:32<00:00, 14.16it/s]\nEpoch 30 finished. Avg Loss: 0.0447","metadata":{}},{"cell_type":"code","source":"class TestImageDataset(Dataset):\n    def __init__(self):\n        super().__init__()\n        self.file = pd.read_csv('/kaggle/input/competitions/digit-recognizer/test.csv')\n        self.pixels = torch.tensor(self.file.values, dtype=torch.float32)\n\n    def __getitem__(self, idx):\n        return self.pixels[idx].unsqueeze(0)\n    def __len__(self):\n        return len(self.pixels)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"test = TestImageDataset()\ntest = DataLoader(test, batch_size=32, shuffle=False)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model.eval()\npredictions = []\n\nwith torch.no_grad():\n    for x_batch in tqdm(test):\n        x_batch = x_batch.view(-1, 1, 28, 28)\n        pred = model(x_batch)\n\n        _, predicted = torch.max(pred, dim=1)\n\n        predictions.extend(predicted.cpu().numpy())","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"100%|██████████| 875/875 [00:10<00:00, 79.97it/s]","metadata":{}},{"cell_type":"code","source":"sub = pd.DataFrame(\n    {\n        'ImageId': np.arange(1, len(predictions)+1),\n        'Label': predictions\n    }\n)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"sub.to_csv('submission.csv', index=False)","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}