\n \ub525\ub7ec\ub2dd(Deep Learning)\uc740 \uc778\uacf5\uc2e0\uacbd\ub9dd\uc744 \uae30\ubc18\uc73c\ub85c \ud558\ub294 \uae30\uacc4\ud559\uc2b5\uc758 \ud55c \ubd84\uc57c\ub85c, \ub2e4\uc591\ud55c \ub370\uc774\ud130\ub85c\ubd80\ud130 \uc720\uc6a9\ud55c \ud328\ud134\uc744 \uc790\ub3d9\uc73c\ub85c \ud559\uc2b5\ud558\ub294 \ub370 \uc8fc\ub825\ud569\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58(PyTorch)\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ucd95\ud558\uace0 \ud559\uc2b5\ud558\ub294 \uacfc\uc815\uc744 \uc0c1\uc138\ud788 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4. \ud559\uc2b5\uc774 \ud544\uc694\ud55c \ub370\uc774\ud130 \ubc0f \ube44\uc988\ub2c8\uc2a4 \uc694\uad6c\uc5d0 \ub530\ub77c \ub2e4\uc591\ud55c \ub124\ud2b8\uc6cc\ud06c \uc544\ud0a4\ud14d\ucc98\ub97c \uc124\uacc4\ud560 \uc218 \uc788\uc73c\uba70, \ud30c\uc774\ud1a0\uce58\ub294 \uc774\uc5d0 \ub9e4\uc6b0 \uc720\uc6a9\ud55c \ub3c4\uad6c\uc785\ub2c8\ub2e4.\n <\/p>\n
1. PyTorch\ub780?<\/h2>\n
\n PyTorch\ub294 \ud398\uc774\uc2a4\ubd81 AI \ub9ac\uc11c\uce58 \uadf8\ub8f9(Facebook AI Research)\uc5d0\uc11c \uac1c\ubc1c\ud55c \uc624\ud508\uc18c\uc2a4 \uba38\uc2e0\ub7ec\ub2dd \ub77c\uc774\ube0c\ub7ec\ub9ac\ub85c, \uc8fc\ub85c \ub525\ub7ec\ub2dd \uc5f0\uad6c \ubc0f \ud504\ub85c\ub355\uc158\uc5d0 \uc0ac\uc6a9\ub429\ub2c8\ub2e4.
\n \ud150\uc11c(Tensor) \uacc4\uc0b0\uacfc \uc790\ub3d9 \ubbf8\ubd84(Autograd) \uae30\ub2a5\uc744 \uc81c\uacf5\ud558\uc5ec \ubaa8\ub378 \ud559\uc2b5 \ubc0f \uadf8\ub798\ub514\uc5b8\ud2b8 \uae30\ubc18 \ucd5c\uc801\ud654\uac00 \uc6a9\uc774\ud569\ub2c8\ub2e4.
\n \ub610\ud55c, Python\uacfc\uc758 \ud1b5\ud569\uc774 \uc798 \ub418\uc5b4 \uc788\uc5b4 \uc9c1\uad00\uc801\uc778 \ucf54\ub4dc \uc791\uc131\uc744 \uc9c0\uc6d0\ud569\ub2c8\ub2e4.\n <\/p>\n
2. PyTorch \uc124\uce58\ud558\uae30<\/h2>\n
\n PyTorch\ub97c \uc124\uce58\ud558\ub294 \ubc29\ubc95\uc740 \uc5ec\ub7ec \uac00\uc9c0\uac00 \uc788\uc73c\uba70, \uc544\ub798\uc758 \uba85\ub839\uc5b4\ub97c \ud1b5\ud574 Conda \ub610\ub294 pip\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \n # Anaconda\ub97c \uc0ac\uc6a9\ud558\ub294 \uacbd\uc6b0\n conda install pytorch torchvision torchaudio cpuonly -c pytorch\n \n # pip\ub97c \uc0ac\uc6a9\ud558\ub294 \uacbd\uc6b0\n pip install torch torchvision torchaudio\n <\/code>\n <\/pre>\n\n \uc124\uce58 \ud6c4, \ub2e4\uc74c\uc758 \ucf54\ub4dc\ub97c \uc2e4\ud589\ud558\uc5ec \uc124\uce58\uac00 \uc815\uc0c1\uc801\uc73c\ub85c \uc644\ub8cc\ub418\uc5c8\ub294\uc9c0 \ud655\uc778\ud574\ubcf4\uc138\uc694.\n <\/p>\n
\n \n import torch\n print(torch.__version__)\n <\/code>\n <\/pre>\n3. \ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8 \uac1c\ub150<\/h2>\n
\n \ub525\ub7ec\ub2dd\uc758 \uc8fc\uc694 \uac1c\ub150\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:\n <\/p>\n
\n- \uc2e0\uacbd\ub9dd(Neural Network):<\/strong> \uc785\ub825\uce35, \uc740\ub2c9\uce35, \ucd9c\ub825\uce35\uc73c\ub85c \uad6c\uc131\ub41c \ub370\uc774\ud130 \ucc98\ub9ac \uad6c\uc870\uc785\ub2c8\ub2e4.<\/li>\n
- \ud150\uc11c(Tensor):<\/strong> \ud30c\uc774\ud1a0\uce58\uc758 \uae30\ubcf8 \ub370\uc774\ud130 \uad6c\uc870\ub85c, \ub2e4\ucc28\uc6d0 \ubc30\uc5f4\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4.<\/li>\n
- \ud65c\uc131\ud654 \ud568\uc218(Activation Function):<\/strong> \uc2e0\uacbd\ub9dd\uc758 \uac01 \ub178\ub4dc\uac00 \ud574\uce58\ub97c \ud1b5\ud574 \ud65c\uc131\ud654\ub418\ub294 \ubc29\uc2dd\uc744 \uacb0\uc815\ud569\ub2c8\ub2e4.<\/li>\n
- \uc190\uc2e4 \ud568\uc218(Loss Function):<\/strong> \ubaa8\ub378\uc758 \uc608\uce21\uacfc \uc2e4\uc81c \uac12 \uc0ac\uc774\uc758 \uc624\ucc28\ub97c \uce21\uc815\ud558\ub294 \ud568\uc218\uc785\ub2c8\ub2e4.<\/li>\n
- \ucd5c\uc801\ud654\uae30(Optimizer):<\/strong> \uc190\uc2e4 \ud568\uc218\ub97c \uae30\ubc18\uc73c\ub85c \ub124\ud2b8\uc6cc\ud06c\uc758 \uac00\uc911\uce58\ub97c \uc5c5\ub370\uc774\ud2b8\ud558\ub294 \uc54c\uace0\ub9ac\uc998\uc785\ub2c8\ub2e4.<\/li>\n<\/ul>\n
4. PyTorch\ub85c \ub525\ub7ec\ub2dd \ubaa8\ub378 \uad6c\ucd95\ud558\uae30<\/h2>\n
\n PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd\uc744 \uad6c\ucd95\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc6b0\ub9ac\uac00 \uc0ac\uc6a9\ud560 \ub370\uc774\ud130\uc14b\uc740 \uc720\uba85\ud55c MNIST \uc190\uae00\uc528 \uc22b\uc790 \ub370\uc774\ud130\uc14b\uc785\ub2c8\ub2e4. \uc774 \ub370\uc774\ud130\uc14b\uc740 0\ubd80\ud130 9\uae4c\uc9c0\uc758 \uc22b\uc790\uac00 \ud3ec\ud568\ub41c \ud751\ubc31 \uc774\ubbf8\uc9c0\ub85c \uad6c\uc131\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
4.1 \ub370\uc774\ud130\uc14b \ub2e4\uc6b4\ub85c\ub4dc<\/h3>\n
\n PyTorch\ub294 torchvision \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud1b5\ud574 \ub2e4\uc591\ud55c \uc774\ubbf8\uc9c0 \ub370\uc774\ud130\uc14b\uc744 \uc27d\uac8c \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0 \uc804\ucc98\ub9ac\ud560 \uc218 \uc788\uac8c \ub3c4\uc640\uc90d\ub2c8\ub2e4.
\n MNIST \ub370\uc774\ud130\uc14b\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0 \ub370\uc774\ud130\ub85c\ub354(DataLoader)\ub97c \uc124\uc815\ud558\ub294 \ucf54\ub4dc\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \n import torch\n from torchvision import datasets, transforms\n from torch.utils.data import DataLoader\n\n # \ub370\uc774\ud130 \uc804\ucc98\ub9ac: \uc774\ubbf8\uc9c0\ub97c \ud150\uc11c\ub85c \ubcc0\ud658\ud558\uace0 \uc815\uaddc\ud654\n transform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,))\n ])\n\n # \ub370\uc774\ud130\uc14b \ub2e4\uc6b4\ub85c\ub4dc\n train_dataset = datasets.MNIST(root='data', train=True, download=True, transform=transform)\n test_dataset = datasets.MNIST(root='data', train=False, download=True, transform=transform)\n\n # \ub370\uc774\ud130\ub85c\ub354 \uc124\uc815\n train_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)\n test_loader = DataLoader(dataset=test_dataset, batch_size=64, shuffle=False)\n <\/code>\n <\/pre>\n4.2 \uc2e0\uacbd\ub9dd \ubaa8\ub378 \uc815\uc758<\/h3>\n
\n \uc774\uc81c \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc815\uc758\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc544\ub798\uc758 \ucf54\ub4dc\ub294 \ub450 \uac1c\uc758 \uc740\ub2c9\uce35\uc744 \uac00\uc9c4 \uc2e0\uacbd\ub9dd\uc744 \ub098\ud0c0\ub0c5\ub2c8\ub2e4.\n <\/p>\n
\n \n import torch.nn as nn\n import torch.nn.functional as F\n\n class SimpleNet(nn.Module):\n def __init__(self):\n super(SimpleNet, self).__init__()\n self.fc1 = nn.Linear(28 * 28, 128) # \uc785\ub825\uce35\n self.fc2 = nn.Linear(128, 64) # \uc740\ub2c9\uce35 1\n self.fc3 = nn.Linear(64, 10) # \ucd9c\ub825\uce35\n\n def forward(self, x):\n x = x.view(-1, 28 * 28) # Flatten\n x = F.relu(self.fc1(x))\n x = F.relu(self.fc2(x))\n x = self.fc3(x)\n return x\n <\/code>\n <\/pre>\n4.3 \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654\uae30 \uc124\uc815<\/h3>\n
\n \ubaa8\ub378\uc744 \ud559\uc2b5\ud558\uae30 \uc704\ud574 \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654\uae30\ub97c \uc815\uc758\ud574\uc57c \ud569\ub2c8\ub2e4. \uc774 \uacbd\uc6b0, \uc190\uc2e4 \ud568\uc218\ub85c\ub294 \uad50\ucc28 \uc5d4\ud2b8\ub85c\ud53c \uc190\uc2e4(Cross Entropy Loss)\uc744 \uc0ac\uc6a9\ud558\uace0, \ucd5c\uc801\ud654\uae30\ub85c\ub294 Adam \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \n model = SimpleNet()\n criterion = nn.CrossEntropyLoss()\n optimizer = torch.optim.Adam(model.parameters(), lr=0.001)\n <\/code>\n <\/pre>\n4.4 \ubaa8\ub378 \ud559\uc2b5<\/h3>\n
\n \uc544\ub798\uc758 \ucf54\ub4dc\ub294 \ubaa8\ub378\uc744 \ud559\uc2b5\ud558\ub294 \uacfc\uc815\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4. \ub370\uc774\ud130 \ub85c\ub354\ub97c \ud1b5\ud574 \ubbf8\ub2c8\ubc30\uce58 \ub2e8\uc704\ub85c \ub370\uc774\ud130\ub97c \uac00\uc838\uc624\uace0, \uac01 \ubc30\uce58\uc5d0 \ub300\ud574 \ubaa8\ub378\uc758 \ucd9c\ub825\uc744 \uacc4\uc0b0\ud55c \ub4a4, \uc190\uc2e4\uc744 \uacc4\uc0b0\ud558\uace0 \uac00\uc911\uce58\ub97c \uc5c5\ub370\uc774\ud2b8\ud569\ub2c8\ub2e4.\n <\/p>\n
\n \n num_epochs = 5\n\n for epoch in range(num_epochs):\n for images, labels in train_loader:\n # \uacbd\ub7c9\ud654: \uae30\uc6b8\uae30 \ucd08\uae30\ud654\n optimizer.zero_grad()\n \n # \uc21c\uc804\ud30c\n outputs = model(images)\n loss = criterion(outputs, labels)\n\n # \uc5ed\uc804\ud30c\n loss.backward()\n optimizer.step()\n\n print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item():.4f}')\n <\/code>\n <\/pre>\n4.5 \ubaa8\ub378 \ud3c9\uac00<\/h3>\n
\n \ubaa8\ub378\uc774 \ud559\uc2b5\ub41c \ud6c4, \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc5d0 \ub300\ud55c \uc815\ud655\ub3c4\ub97c \uce21\uc815\ud558\ub294 \ubc29\ubc95\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4.\n <\/p>\n
\n \n model.eval() # \ud3c9\uac00 \ubaa8\ub4dc\ub85c \uc804\ud658\n correct = 0\n total = 0\n\n with torch.no_grad(): # \uae30\uc6b8\uae30 \uacc4\uc0b0 \ube44\ud65c\uc131\ud654\n for images, labels in test_loader:\n outputs = model(images)\n _, predicted = torch.max(outputs.data, 1)\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n\n accuracy = 100 * correct \/ total\n print(f'Accuracy on the test set: {accuracy:.2f}%')\n <\/code>\n <\/pre>\n5. \ub525\ub7ec\ub2dd\uc5d0\uc11c\uc758 \ud558\uc774\ud37c\ud30c\ub77c\ubbf8\ud130 \ud29c\ub2dd<\/h2>\n
\n \ud558\uc774\ud37c\ud30c\ub77c\ubbf8\ud130 \ud29c\ub2dd\uc740 \ub525\ub7ec\ub2dd \ubaa8\ub378 \uc131\ub2a5\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\ub294 \ub370 \uc911\uc694\ud55c \ub2e8\uacc4\uc785\ub2c8\ub2e4. \ud558\uc774\ud37c\ud30c\ub77c\ubbf8\ud130\uc5d0\ub294 \ud559\uc2b5\ub960, \ubc30\uce58 \ud06c\uae30, \uc740\ub2c9\uce35\uc758 \ud06c\uae30 \ubc0f \uc218, \ud65c\uc131\ud654 \ud568\uc218 \uc720\ud615, \ub4dc\ub86d\uc544\uc6c3 \ube44\uc728 \ub4f1\uc774 \ud3ec\ud568\ub429\ub2c8\ub2e4.\n <\/p>\n
\n \uc77c\ubc18\uc801\uc73c\ub85c \ud558\uc774\ud37c\ud30c\ub77c\ubbf8\ud130 \ud29c\ub2dd\uc744 \uc704\ud574 Grid Search, Random Search, Bayesian Optimization \ub4f1\uc758 \uae30\ubc95\uc774 \uc0ac\uc6a9\ub418\uba70, \uac01 \ubc29\ubc95\uc740 \ub2e4\uc591\ud55c \uc870\ud569\uc744 \ud3c9\uac00\ud558\uc5ec \ucd5c\uc801\uc758 \uc124\uc815\uc744 \ud0d0\uc0c9\ud569\ub2c8\ub2e4.\n <\/p>\n
6. \uacb0\ub860<\/h2>\n
\n \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec \uae30\ubcf8\uc801\uc778 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ucd95\ud558\uace0 \ud559\uc2b5\ud558\ub294 \uacfc\uc815\uc744 \uc18c\uac1c\ud558\uc600\uc2b5\ub2c8\ub2e4. \ub370\uc774\ud130\uc14b \uc900\ube44, \ubaa8\ub378 \uc815\uc758, \ud559\uc2b5, \ud3c9\uac00 \ub4f1 \ub525\ub7ec\ub2dd\uc758 \uc8fc\uc694 \ub2e8\uacc4\ub97c \ub2e4\ub8e8\uc5c8\uc2b5\ub2c8\ub2e4.
\n \ub525\ub7ec\ub2dd\uc758 \uc774\ud574\ub97c \ub3d5\uae30 \uc704\ud574 \ub2e4\uc591\ud55c \uc774\ub860 \ubc0f \uae30\ubc95\ub3c4 \uc124\uba85\ud558\uc600\uc73c\ub2c8, \uc774\ub97c \ubc14\ud0d5\uc73c\ub85c \ub354 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uacfc \ub2e4\uc591\ud55c \uc751\uc6a9 \ubd84\uc57c\uc5d0 \ub3c4\uc804\ud574 \ubcf4\uc2dc\uae30 \ubc14\ub78d\ub2c8\ub2e4.\n <\/p>\n
7. \ucc38\uace0 \uc790\ub8cc<\/h2>\n