\n \ub525\ub7ec\ub2dd\uc740 \ucd5c\uadfc \uba87 \ub144 \ub3d9\uc548 \uc778\uacf5\uc9c0\ub2a5 \ubc0f \uba38\uc2e0\ub7ec\ub2dd \ubd84\uc57c\uc5d0\uc11c \uc6b0\uc138\ud55c \ubc29\ubc95\ub860\uc73c\ub85c \uc790\ub9ac \uc7a1\uc558\uc2b5\ub2c8\ub2e4. \uc624\ub298\uc740 \uadf8 \uc911\uc5d0\uc11c\ub3c4 \ud569\uc131\uacf1 \uc2e0\uacbd\ub9dd(Convolutional Neural Network, CNN)\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. CNN\uc740 \uc774\ubbf8\uc9c0 \uc778\uc2dd \ubc0f \ucc98\ub9ac\uc5d0 \ud2b9\ud788 \ud6a8\uacfc\uc801\uc774\uba70, \uc5ec\ub7ec \uc0b0\uc5c5\uc5d0\uc11c \uad11\ubc94\uc704\ud558\uac8c \uc0ac\uc6a9\ub429\ub2c8\ub2e4.\n<\/p>\n
\n \ud569\uc131\uacf1 \uc2e0\uacbd\ub9dd\uc740 \ubc14\ub85c \uc0ac\uc9c4\uc774\ub098 \ube44\ub514\uc624 \uac19\uc740 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\uc758 \uc2dc\uac01\uc801 \ud328\ud134\uc744 \uc778\uc2dd\ud558\ub294 \ub370 \ud2b9\ud654\ub41c \uc2e0\uacbd\ub9dd\uc758 \uc77c\uc885\uc785\ub2c8\ub2e4. CNN\uc740 \uae30\ubcf8\uc801\uc73c\ub85c \ud569\uc131\uacf1 \uce35, \ud480\ub9c1 \uce35, \uc644\uc804 \uc5f0\uacb0 \uce35\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4.\n<\/p>\n
\n \ud569\uc131\uacf1 \uce35\uc740 \uc785\ub825 \ub370\uc774\ud130\uc5d0\uc11c \ud2b9\uc9d5\uc744 \ucd94\ucd9c\ud558\ub294 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. \uc774 \uce35\uc740 \uc791\uc740 \ud544\ud130(\ucee4\ub110)\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc785\ub825 \uc774\ubbf8\uc9c0\uc758 \ud2b9\uc815 \ubd80\ubd84\uacfc \uc5f0\uc0b0\uc744 \uc218\ud589\ud558\uc5ec \ucd9c\ub825\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4. \uc774\ub807\uac8c \uc0dd\uc131\ub41c \ud2b9\uc131 \ub9f5(feature map)\uc740 \uc785\ub825 \ub370\uc774\ud130\uc758 \uc720\uc6a9\ud55c \uc815\ubcf4\ub9cc\uc744 \ub2f4\uace0 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \ud480\ub9c1 \uce35\uc740 \ud2b9\uc131 \ub9f5\uc758 \ud06c\uae30\ub97c \uc904\uc774\ub294 \ub370 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc774\ub294 \ubaa8\ub378\uc758 \ubcf5\uc7a1\uc131\uc744 \uc904\uc774\uace0 \uc5f0\uc0b0\ub7c9\uc744 \uac10\uc18c\uc2dc\ucf1c \uacfc\uc801\ud569\uc744 \ubc29\uc9c0\ud558\ub294 \ub370 \ub3c4\uc6c0\uc744 \uc90d\ub2c8\ub2e4. \uac00\uc7a5 \uc77c\ubc18\uc801\uc778 \ubc29\ubc95\uc740 \ucd5c\ub300 \ud480\ub9c1(max pooling)\uc73c\ub85c, \uc774 \ubc29\ubc95\uc740 \uc8fc\uc5b4\uc9c4 \uc601\uc5ed\uc5d0\uc11c \uac00\uc7a5 \ud070 \uac12\uc744 \uc120\ud0dd\ud558\uc5ec \ud2b9\uc131 \ub9f5\uc758 \ud06c\uae30\ub97c \uc904\uc785\ub2c8\ub2e4.\n<\/p>\n
\n \uc2e0\uacbd\ub9dd\uc758 \ub9c8\uc9c0\ub9c9 \ubd80\ubd84\uc5d0\uc11c\ub294 \uc644\uc804 \uc5f0\uacb0 \uce35\uc774 \uc874\uc7ac\ud569\ub2c8\ub2e4. \uc774 \uce35\uc740 \uc774\uc804 \uce35\uc5d0\uc11c \uc5bb\uc740 \uc815\ubcf4\ub4e4\uc744 \ubc14\ud0d5\uc73c\ub85c \ucd5c\uc885 \uc608\uce21\uc744 \ud558\uac8c \ub429\ub2c8\ub2e4. \ubaa8\ub4e0 \ub274\ub7f0\uc774 \uc774\uc804 \uce35\uacfc \uc5f0\uacb0\ub418\uc5b4 \uc788\ub294 \uad6c\uc870\uc774\ubbc0\ub85c, \uc785\ub825 \ub370\uc774\ud130\uc5d0 \ub300\ud55c \ubcf5\uc7a1\ud55c \uacb0\uc815\uc744 \ub0b4\ub9b4 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \uc774\uc81c \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac04\ub2e8\ud55c CNN \ubaa8\ub378\uc744 \uad6c\ud604\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. MNIST \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc190\uae00\uc528 \uc22b\uc790\ub97c \ubd84\ub958\ud558\ub294 \ubaa8\ub378\uc744 \ub9cc\ub4e4\uc5b4 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \uba3c\uc800 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud558\uace0 \ub370\uc774\ud130\uc14b\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uaca0\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\uc785\ub2c8\ub2e4:\n<\/p>\n
pip install torch torchvision\n<\/code><\/pre>\n\ub370\uc774\ud130\uc14b \uc900\ube44<\/h3>\n
\n MNIST \ub370\uc774\ud130\uc14b\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0 \ubd88\ub7ec\uc635\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud559\uc2b5 \ubc0f \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc744 \uc900\ube44\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\nimport torch\nimport torchvision\nimport torchvision.transforms as transforms\n\n# \ub370\uc774\ud130 \ubcc0\ud658 \uc815\uc758\ntransform = transforms.Compose(\n [transforms.ToTensor(), \n transforms.Normalize((0.5,), (0.5,))]) # \ud3c9\uade0\uacfc \ud45c\uc900\ud3b8\ucc28\ub97c \uc0ac\uc6a9\ud55c \uc815\uaddc\ud654\n\n# \ud559\uc2b5 \ub370\uc774\ud130\uc14b\ntrainset = torchvision.datasets.MNIST(root='.\/data', train=True, download=True, transform=transform)\ntrainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)\n\n# \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\ntestset = torchvision.datasets.MNIST(root='.\/data', train=False, download=True, transform=transform)\ntestloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=False)\n<\/code><\/pre>\n\ubaa8\ub378 \uc815\uc758<\/h3>\n
\n \uc774\uc81c \ud569\uc131\uacf1 \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc815\uc758\ud574 \ubd05\uc2dc\ub2e4. CNN \ubaa8\ub378\uc740 \uc77c\ubc18\uc801\uc73c\ub85c \ud569\uc131\uacf1 \uce35\uacfc \ud480\ub9c1 \uce35\uc744 \uc870\ud569\ud55c \uad6c\uc870\ub85c \uc124\uacc4\ub429\ub2c8\ub2e4.\n<\/p>\n
\nimport torch.nn as nn\nimport torch.nn.functional as F\n\nclass CNN(nn.Module):\n def __init__(self):\n super(CNN, self).__init__()\n self.conv1 = nn.Conv2d(1, 32, kernel_size=3) # \uc785\ub825 \ucc44\ub110 1, \ucd9c\ub825 \ucc44\ub110 32\n self.pool = nn.MaxPool2d(kernel_size=2, stride=2) # \ucd5c\ub300 \ud480\ub9c1\n self.conv2 = nn.Conv2d(32, 64, kernel_size=3) # \uc785\ub825 \ucc44\ub110 32, \ucd9c\ub825 \ucc44\ub110 64\n self.fc1 = nn.Linear(64 * 6 * 6, 128) # \uc644\uc804 \uc5f0\uacb0 \uce35, 64x6x6\uc740 \ucd9c\ub825 \ud06c\uae30\n self.fc2 = nn.Linear(128, 10) # \ucd5c\uc885 \ucd9c\ub825 10\uac1c(0~9)\n\n def forward(self, x):\n x = self.pool(F.relu(self.conv1(x))) # \uccab \ubc88\uc9f8 \ud569\uc131\uacf1\uacfc \ud480\ub9c1\n x = self.pool(F.relu(self.conv2(x))) # \ub450 \ubc88\uc9f8 \ud569\uc131\uacf1\uacfc \ud480\ub9c1\n x = x.view(-1, 64 * 6 * 6) # \ud150\uc11c \ud615\ud0dc \ubcc0\uacbd\n x = F.relu(self.fc1(x)) # \uccab \ubc88\uc9f8 \uc644\uc804 \uc5f0\uacb0 \uce35\n x = self.fc2(x) # \ub450 \ubc88\uc9f8 \uc644\uc804 \uc5f0\uacb0 \uce35\n return x\n<\/code><\/pre>\n\ubaa8\ub378 \ud559\uc2b5<\/h3>\n
\n \ubaa8\ub378\uc744 \ud559\uc2b5\uc2dc\ud0a4\uae30 \uc704\ud574\uc11c\ub294 \uc190\uc2e4 \ud568\uc218\uc640 \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc815\uc758\ud558\uace0 \ub370\uc774\ud130\uc5d0 \ub300\ud574 \ubc18\ubcf5\uc801\uc73c\ub85c \ud559\uc2b5\uc744 \uc9c4\ud589\ud574\uc57c \ud569\ub2c8\ub2e4.\n<\/p>\n
\n# \ubaa8\ub378 \ucd08\uae30\ud654\ncnn = CNN()\ncriterion = nn.CrossEntropyLoss() # \uc190\uc2e4 \ud568\uc218\noptimizer = torch.optim.SGD(cnn.parameters(), lr=0.01) # \ud655\ub960\uc801 \uacbd\uc0ac \ud558\uac15\ubc95\n\n# \ubaa8\ub378 \ud559\uc2b5\nfor epoch in range(5): # \uc5d0\ud3ed \uc218\n running_loss = 0.0\n for i, data in enumerate(trainloader, 0):\n inputs, labels = data\n optimizer.zero_grad() # \uae30\uc6b8\uae30 \ucd08\uae30\ud654\n outputs = cnn(inputs) # \uc608\uce21\n loss = criterion(outputs, labels) # \uc190\uc2e4 \uacc4\uc0b0\n loss.backward() # \uae30\uc6b8\uae30 \uacc4\uc0b0\n optimizer.step() # \ud30c\ub77c\ubbf8\ud130 \uc5c5\ub370\uc774\ud2b8\n running_loss += loss.item()\n print(f'Epoch {epoch + 1}, Loss: {running_loss \/ len(trainloader)}')\n<\/code><\/pre>\n\ubaa8\ub378 \ud3c9\uac00<\/h3>\n
\n \ud559\uc2b5\uc774 \uc644\ub8cc\ub41c \ud6c4, \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc744 \ud1b5\ud574 \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4.\n<\/p>\n
\ncorrect = 0\ntotal = 0\n\nwith torch.no_grad(): # \uae30\uc6b8\uae30 \uacc4\uc0b0 \ube44\ud65c\uc131\ud654\n for data in testloader:\n images, labels = data\n outputs = cnn(images)\n _, predicted = torch.max(outputs.data, 1)\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n\nprint(f'Accuracy on the test set: {100 * correct \/ total} %')\n<\/code><\/pre>\n\uacb0\ub860<\/h2>\n
\n \uc774\uc81c \uc6b0\ub9ac\ub294 \uac04\ub2e8\ud55c \ud569\uc131\uacf1 \uc2e0\uacbd\ub9dd\uc744 \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec \uad6c\uc131\ud558\uace0, \uc2e4\uc81c \ub370\uc774\ud130\uc14b\uc744 \ud1b5\ud574 \ud559\uc2b5\uc2dc\ud0a4\uace0 \ud3c9\uac00\ud558\ub294 \uacfc\uc815\uc744 \uacbd\ud5d8\ud588\uc2b5\ub2c8\ub2e4. \uc774 \uac15\uc88c\ub97c \ud1b5\ud574 \ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8\uc801\uc778 \uad6c\uc870\uc640 \ud30c\uc774\uc36c\uc744 \ud65c\uc6a9\ud55c \uc2e4\uc2b5\uc744 \uc774\ud574\ud558\ub294 \ub370 \ub3c4\uc6c0\uc774 \ub418\uc168\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4. \uc55e\uc73c\ub85c \ub354 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uacfc \ub2e4\uc591\ud55c \ub370\uc774\ud130\uc14b\uc744 \ub2e4\ub8e8\ub294 \ub370\uc5d0\ub3c4 \ub3c4\uc804\ud574\ubcf4\uc2ed\uc2dc\uc624!\n<\/p>\n
\ucc38\uace0\uc790\ub8cc<\/h2>\n