\ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8 \uad6c\uc131 \uc694\uc18c \uc911 \ud558\ub098\ub294 \ud150\uc11c(tensor)\uc785\ub2c8\ub2e4. \ud150\uc11c\ub294 N\ucc28\uc6d0 \ubc30\uc5f4\uc744 \ub098\ud0c0\ub0b4\uba70, \ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\ub294 \uc2e0\uacbd\ub9dd \ud559\uc2b5\uc758 \uae30\ucd08\ub85c \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\uc5d0\uc11c \ud150\uc11c\ub97c \uc0dd\uc131\ud558\uace0 \ub2e4\ub8e8\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\ud150\uc11c\ub294 \uae30\ubcf8\uc801\uc73c\ub85c \uc22b\uc790\uc758 \uc9d1\ud569\uc785\ub2c8\ub2e4. 0\ucc28\uc6d0 \ud150\uc11c\ub294 \uc2a4\uce7c\ub77c(scalar), 1\ucc28\uc6d0 \ud150\uc11c\ub294 \ubca1\ud130(vector), 2\ucc28\uc6d0 \ud150\uc11c\ub294 \ud589\ub82c(matrix), 3\ucc28\uc6d0 \ud150\uc11c\ub294 \ub2e4\ucc28\uc6d0 \ubc30\uc5f4(array)\uc73c\ub85c \ubd88\ub9bd\ub2c8\ub2e4. \ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\ub294 \ud150\uc11c\ub97c \uc27d\uac8c \uc0dd\uc131\ud558\uace0 \uc870\uc791\ud560 \uc218 \uc788\ub294 \ub2e4\uc591\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4.<\/p>\n
\uba3c\uc800, \ud30c\uc774\ud1a0\uce58\ub97c \uc124\uce58\ud574\uc57c \ud569\ub2c8\ub2e4. \uc544\ub098\ucf58\ub2e4\ub97c \uc0ac\uc6a9\ud55c\ub2e4\uba74 \uc544\ub798 \ucf54\ub4dc\ub97c \uc2e4\ud589\ud558\uc5ec \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
conda install pytorch torchvision torchaudio cpuonly -c pytorch<\/code><\/pre>\n2. \ud150\uc11c \uc0dd\uc131\ud558\uae30<\/h2>\n
\ud30c\uc774\ud1a0\uce58\uc5d0\uc11c \ud150\uc11c\ub97c \uc0dd\uc131\ud558\ub294 \ubc29\ubc95\uc740 \uc5ec\ub7ec \uac00\uc9c0\uc785\ub2c8\ub2e4. \uac00\uc7a5 \uae30\ubcf8\uc801\uc778 \ubc29\ubc95\uc740 torch.tensor()<\/code> \ud568\uc218\ub97c \uc0ac\uc6a9\ud558\ub294 \uac83\uc785\ub2c8\ub2e4.<\/p>\n2.1. \uae30\ubcf8\uc801\uc778 \ud150\uc11c \uc0dd\uc131<\/h3>\nimport torch\n\n# \ub9ac\uc2a4\ud2b8\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud150\uc11c \uc0dd\uc131\ntensor1 = torch.tensor([1, 2, 3])\nprint(tensor1)<\/code><\/pre>\n\uc704 \ucf54\ub4dc\ub97c \uc2e4\ud589\ud558\uba74 \ub2e4\uc74c\uacfc \uac19\uc740 \uacb0\uacfc\ub97c \uc5bb\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
tensor([1, 2, 3])<\/code><\/pre>\n2.2. \ub2e4\uc591\ud55c \ud150\uc11c \uc0dd\uc131 \ubc29\ubc95<\/h3>\n
\ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\ub294 \ub2e4\uc591\ud55c \ubc29\ubc95\uc73c\ub85c \ud150\uc11c\ub97c \uc0dd\uc131\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4:<\/p>\n
\ntorch.zeros()<\/code>: \ubaa8\ub4e0 \uc694\uc18c\uac00 0\uc778 \ud150\uc11c \uc0dd\uc131<\/li>\ntorch.ones()<\/code>: \ubaa8\ub4e0 \uc694\uc18c\uac00 1\uc778 \ud150\uc11c \uc0dd\uc131<\/li>\ntorch.arange()<\/code>: \uc9c0\uc815\ud55c \ubc94\uc704\uc758 \uc694\uc18c\ub97c \uac00\uc9c4 \ud150\uc11c \uc0dd\uc131<\/li>\ntorch.randn()<\/code>: \ud3c9\uade0 0, \ud45c\uc900\ud3b8\ucc28 1\uc758 \uc815\uaddc\ubd84\ud3ec\ub97c \ub530\ub974\ub294 \ud150\uc11c \uc0dd\uc131<\/li>\n<\/ul>\n\uc608\uc81c \ucf54\ub4dc<\/h4>\n# \ub2e4\uc591\ud55c \ud150\uc11c \uc0dd\uc131\nzeros_tensor = torch.zeros(3, 4)\nones_tensor = torch.ones(3, 4)\narange_tensor = torch.arange(0, 10, step=1)\nrandom_tensor = torch.randn(3, 4)\n\nprint(\"Zeros Tensor:\\n\", zeros_tensor)\nprint(\"Ones Tensor:\\n\", ones_tensor)\nprint(\"Arange Tensor:\\n\", arange_tensor)\nprint(\"Random Tensor:\\n\", random_tensor)<\/code><\/pre>\n3. \ud150\uc11c\uc758 \uc18d\uc131<\/h2>\n
\ud150\uc11c\ub294 \ub2e4\uc591\ud55c \uc18d\uc131\uc744 \uac00\uc9d1\ub2c8\ub2e4. \ud150\uc11c\ub97c \uc0dd\uc131\ud55c \ud6c4, \uadf8 \uc18d\uc131\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\ub294 \uc8fc\uc694 \uc18d\uc131\uc785\ub2c8\ub2e4:<\/p>\n
\ntensor.shape<\/code>: \ud150\uc11c\uc758 \ucc28\uc6d0(shape)<\/li>\ntensor.dtype<\/code>: \ud150\uc11c\uc758 \ub370\uc774\ud130 \ud0c0\uc785<\/li>\ntensor.device<\/code>: \ud150\uc11c\uac00 \uc874\uc7ac\ud558\ub294 \ub514\ubc14\uc774\uc2a4(CPU \ub610\ub294 GPU)<\/li>\n<\/ul>\n\uc608\uc81c \ucf54\ub4dc<\/h3>\nprint(\"Shape:\", tensor1.shape)\nprint(\"Data Type:\", tensor1.dtype)\nprint(\"Device:\", tensor1.device)<\/code><\/pre>\n4. \ud150\uc11c \uc5f0\uc0b0<\/h2>\n
\ud150\uc11c\ub294 \ub2e4\uc591\ud55c \uc5f0\uc0b0\uc744 \uc9c0\uc6d0\ud569\ub2c8\ub2e4. \uae30\ubcf8\uc801\uc778 \uc0b0\uc220 \uc5f0\uc0b0\ubd80\ud130, \uace0\uae09 \uc5f0\uc0b0\uae4c\uc9c0 \ub2e4\uc591\ud569\ub2c8\ub2e4.<\/p>\n
4.1. \uae30\ubcf8\uc801\uc778 \uc0b0\uc220 \uc5f0\uc0b0<\/h3>\ntensor_a = torch.tensor([1, 2, 3])\ntensor_b = torch.tensor([4, 5, 6])\n\n# \ub367\uc148\nadd_result = tensor_a + tensor_b\nprint(\"Addition Result:\", add_result)\n\n# \uacf1\uc148\nmul_result = tensor_a * tensor_b\nprint(\"Multiplication Result:\", mul_result)<\/code><\/pre>\n4.2. \ud589\ub82c \uc5f0\uc0b0<\/h3>\n
\ud589\ub82c \uacf1\uc148\uc740 torch.mm()<\/code> \ub610\ub294 @<\/code> \uc5f0\uc0b0\uc790\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\nmatrix_a = torch.tensor([[1, 2],\n [3, 4]])\n\nmatrix_b = torch.tensor([[5, 6],\n [7, 8]])\n\nmatrix_product = torch.mm(matrix_a, matrix_b)\nprint(\"Matrix Product:\\n\", matrix_product)<\/code><\/pre>\n5. \ud150\uc11c \uc2ac\ub77c\uc774\uc2f1 \ubc0f \uc778\ub371\uc2f1<\/h2>\n
\ud150\uc11c\ub294 N\ucc28\uc6d0 \ubc30\uc5f4\uc774\uae30 \ub54c\ubb38\uc5d0 \uc2ac\ub77c\uc774\uc2f1\uacfc \uc778\ub371\uc2f1\uc744 \ud1b5\ud574 \uc6d0\ud558\ub294 \ub370\uc774\ud130\ub97c \ucd94\ucd9c\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
5.1. \uae30\ubcf8 \uc778\ub371\uc2f1<\/h3>\ntensor = torch.tensor([[1, 2, 3],\n [4, 5, 6],\n [7, 8, 9]])\n\n# \uccab \ubc88\uc9f8 \ud589\uacfc \ub450 \ubc88\uc9f8 \uc5f4\uc758 \uc694\uc18c\nelement = tensor[0, 1]\nprint(\"Element at (0, 1):\", element)<\/code><\/pre>\n5.2. \uc2ac\ub77c\uc774\uc2f1<\/h3>\n<\/code#><\/pre>\n6. \ud150\uc11c\uc640 GPU<\/h2>\n
\ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\ub294 GPU\ub97c \ud65c\uc6a9\ud558\uc5ec \uc5f0\uc0b0\uc744 \uac00\uc18d\ud654 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud150\uc11c\ub97c GPU\ub85c \uc774\ub3d9\ud558\ub824\uba74 .to()<\/code> \u65b9\u6cd5\ub97c \uc0ac\uc6a9\ud558\uba74 \ub429\ub2c8\ub2e4.<\/p>\n\uc608\uc81c \ucf54\ub4dc<\/h3>\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\ntensor_gpu = tensor.to(device)\nprint(\"Tensor on GPU:\", tensor_gpu)<\/code><\/pre>\n7. \ud150\uc11c\uc758 \ubcc0\ud615<\/h2>\n
\ud150\uc11c\ub97c \ub2e4\ub8e8\ub2e4 \ubcf4\uba74 \ubaa8\uc591\uc744 \ubcc0\ud615\ud574\uc57c \ud560 \ub54c\uac00 \ub9ce\uc2b5\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 torch.view()<\/code> \ub098 torch.reshape()<\/code>\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n\uc608\uc81c \ucf54\ub4dc<\/h3>\nreshaped_tensor = tensor.view(1, 9)\nprint(\"Reshaped Tensor:\\n\", reshaped_tensor)<\/code><\/pre>\n8. \uc885\ud569 \uc608\uc81c<\/h2>\n
\uc9c0\uae08\uae4c\uc9c0 \ubc30\uc6b4 \ub0b4\uc6a9\ub4e4\uc744 \uc885\ud569\ud558\uc5ec, \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \ub9cc\ub4e4\uc5b4 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. MNIST \ub370\uc774\ud130\uc14b\uc744 \uc774\uc6a9\ud558\uc5ec \uc190\uae00\uc528 \uc22b\uc790\ub97c \ubd84\ub958\ud558\ub294 \ubaa8\ub378\uc744 \ub9cc\ub4e4\uc5b4 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\ud30c\uc774\ud1a0\uce58 \ubaa8\ub378 \uc0dd\uc131<\/h3>\nimport torchvision.transforms as transforms\nfrom torchvision import datasets\nfrom torch.utils.data import DataLoader\n\n# \ub370\uc774\ud130\uc14b \ub2e4\uc6b4\ub85c\ub4dc \ubc0f \ub85c\ub4dc\ntransform = transforms.Compose([transforms.ToTensor()])\ntrain_dataset = datasets.MNIST(root='.\/data', train=True, download=True, transform=transform)\ntrain_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)\n\n# \uac80\uc99d\ud558\uae30 \uc704\ud55c \uac04\ub2e8\ud55c \ubaa8\ub378 \uc815\uc758\nimport torch.nn as nn\nimport torch.optim as optim\n\nclass SimpleNN(nn.Module):\n def __init__(self):\n super(SimpleNN, self).__init__()\n self.fc1 = nn.Linear(28*28, 128) # 28x28\uc740 MNIST \uc774\ubbf8\uc9c0 \ud06c\uae30\n self.fc2 = nn.Linear(128, 10) # 10\uc740 \ubd84\ub958\ud560 \ud074\ub798\uc2a4 \uc218\n\n def forward(self, x):\n x = x.view(-1, 28*28) # flatten the input\n x = torch.relu(self.fc1(x))\n x = self.fc2(x)\n return x\n\n# \ubaa8\ub378\uacfc \uc190\uc2e4 \ud568\uc218 \ubc0f \uc635\ud2f0\ub9c8\uc774\uc800 \uc124\uc815\nmodel = SimpleNN().to(device)\ncriterion = nn.CrossEntropyLoss()\noptimizer = optim.SGD(model.parameters(), lr=0.01)\n\n# \ud6c8\ub828 \ub8e8\ud504\nfor epoch in range(5): # 5 \uc5d0\ud3ed \ud6c8\ub828\n for batch_idx, (data, target) in enumerate(train_loader):\n data, target = data.to(device), target.to(device)\n\n optimizer.zero_grad() # \uc774\uc804 \uadf8\ub798\ub514\uc5b8\ud2b8 0\uc73c\ub85c \ucd08\uae30\ud654\n output = model(data) # \ubaa8\ub378\uc5d0 \ub370\uc774\ud130 \ud1b5\uacfc\n loss = criterion(output, target) # \uc190\uc2e4 \uacc4\uc0b0\n loss.backward() # \uadf8\ub798\ub514\uc5b8\ud2b8 \uacc4\uc0b0\n optimizer.step() # \ud30c\ub77c\ubbf8\ud130 \uc5c5\ub370\uc774\ud2b8\n\n if batch_idx % 100 == 0:\n print(f'Epoch: {epoch}, Batch: {batch_idx}, Loss: {loss.item()}')<\/code><\/pre>\n\uc704\uc758 \ucf54\ub4dc\uc5d0\uc11c \uc6b0\ub9ac\ub294 MNIST \ub370\uc774\ud130\uc14b\uc744 \ud559\uc2b5\ud558\uc5ec \uc190\uae00\uc528 \uc22b\uc790\ub97c \ubd84\ub958\ud558\ub294 \ub2e8\uc21c\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uad6c\ucd95\ud588\uc2b5\ub2c8\ub2e4. \ud150\uc11c\ub97c \uc0dd\uc131\ud558\uace0, \uc5f0\uc0b0\ud558\uace0, GPU\uc5d0\uc11c \uc2e4\ud589\ud558\ub294 \uacfc\uc815\uc744 \ubaa8\ub450 \ud3ec\ud568\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\uc5d0\uc11c \ud150\uc11c\ub97c \uc0dd\uc131\ud558\uace0 \ub2e4\ub8e8\ub294 \ub2e4\uc591\ud55c \ubc29\ubc95\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \ud150\uc11c\ub294 \ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8\uc801\uc778 \uad6c\uc131 \uc694\uc18c\ub85c, \ubaa8\ub378 \ud6c8\ub828 \ubc0f \ud14c\uc2a4\ud2b8 \uacfc\uc815\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. \ub2e4\uc74c \ub2e8\uacc4\uc5d0\uc11c\ub294 \ub354 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uacfc \ub525\ub7ec\ub2dd \uae30\ubc95\uc5d0 \ub300\ud574 \ud559\uc2b5\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uac10\uc0ac\ud569\ub2c8\ub2e4!<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"