\n1. \uc11c\ub860<\/h2>\n
\n \ub525\ub7ec\ub2dd\uc740 \uba38\uc2e0\ub7ec\ub2dd\uc758 \ud55c \ubd84\uc57c\ub85c\uc11c, \uc778\uacf5\uc9c0\ub2a5\uc758 \ud601\uc2e0\uc744 \uc774\ub04c\uace0 \uc788\ub294 \uc911\uc694\ud55c \uae30\uc220\uc785\ub2c8\ub2e4.
\n \ucd5c\uadfc \uba87 \ub144 \uc0ac\uc774 \ud30c\uc774\ud1a0\uce58(PyTorch)\ub294 \uc5f0\uad6c \ubc0f \uac1c\ubc1c\uc5d0 \ub110\ub9ac \uc0ac\uc6a9\ub418\uace0 \uc788\ub294 \ub525\ub7ec\ub2dd \ud504\ub808\uc784\uc6cc\ud06c \uc911 \ud558\ub098\ub85c \uc790\ub9ac \uc7a1\uc558\uc2b5\ub2c8\ub2e4.
\n \uc774 \uae00\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \ubb38\ubc95\uc744 \uc790\uc138\ud788 \uc0b4\ud3b4\ubcf4\uba70, \uc2e4\uc6a9\uc801\uc778 \uc608\uc81c \ucf54\ub4dc\ub97c \ud1b5\ud574 \uc774\ud574\ub97c \ub3d5\uace0\uc790 \ud569\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n2. \ud30c\uc774\ud1a0\uce58 \uc18c\uac1c<\/h2>\n
\n \ud30c\uc774\ud1a0\uce58\ub294 \uc0ac\uc6a9\ud558\uae30 \uc26c\uc6b4 \uc778\ud130\ud398\uc774\uc2a4\ub97c \uc81c\uacf5\ud558\uba70, \ub3d9\uc801 \uacc4\uc0b0 \uadf8\ub798\ud504(dynamically computed graphs)\ub97c \uc9c0\uc6d0\ud569\ub2c8\ub2e4. \uc774\ub294 \uc5f0\uad6c\uc790\ub4e4\uc774 \uc2e4\ud5d8\ud558\uace0 \ubaa8\ub378\uc744 \ubcc0\uacbd\ud558\ub294\ub370 \uc720\uc5f0\uc131\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4.
\n \ub610\ud55c, \ud30c\uc774\ud1a0\uce58\ub294 GPU \uac00\uc18d\uc744 \ud1b5\ud574 \uacc4\uc0b0 \uc18d\ub3c4\ub97c \ub192\uc77c \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc7a5\uc810\ub4e4\ub85c \uc778\ud574 \ud30c\uc774\ud1a0\uce58\ub294 \ub9ce\uc740 \ub370\uc774\ud130 \uacfc\ud559\uc790\uc640 \uc5f0\uad6c\uc790\ub4e4 \uc0ac\uc774\uc5d0\uc11c \uc778\uae30\ub97c \ub04c\uace0 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n3. \ud30c\uc774\ud1a0\uce58 \uc124\uce58<\/h2>\n
\n \ud30c\uc774\ud1a0\uce58\ub97c \uc124\uce58\ud558\uae30 \uc704\ud574\uc11c\ub294 \ud30c\uc774\uc36c\uc774 \uc124\uce58\ub418\uc5b4 \uc788\uc5b4\uc57c \ud569\ub2c8\ub2e4.
\n \ub2e4\uc74c\uacfc \uac19\uc740 \uba85\ub839\uc5b4\ub85c \ud30c\uc774\ud1a0\uce58\ub97c \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:\n <\/p>\n
\n pip install torch torchvision torchaudio<\/code>\n <\/pre>\n<\/section>\n\n4. \uae30\ubcf8 \ubb38\ubc95<\/h2>\n4.1 Tensor<\/h3>\n
\n \ud30c\uc774\ud1a0\uce58\uc5d0\uc11c \uac00\uc7a5 \uae30\ubcf8\uc801\uc778 \ub370\uc774\ud130 \uad6c\uc870\ub294 Tensor\uc785\ub2c8\ub2e4. Tensor\ub294 \ub2e4\ucc28\uc6d0 \ubc30\uc5f4\ub85c,
\n NumPy \ubc30\uc5f4\uacfc \uc720\uc0ac\ud558\uc9c0\ub9cc GPU\uc5d0\uc11c \uacc4\uc0b0\uc744 \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.
\n Tensor\ub97c \uc0dd\uc131\ud558\ub294 \ubc29\ubc95 \uba87 \uac00\uc9c0\ub97c \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \nimport torch\n\n# 1D Tensor \uc0dd\uc131\ntensor_1d = torch.tensor([1.0, 2.0, 3.0])\nprint(tensor_1d)\n\n# 2D Tensor \uc0dd\uc131\ntensor_2d = torch.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])\nprint(tensor_2d)\n\n# \ub79c\ub364 Tensor \uc0dd\uc131\nrandom_tensor = torch.randn(2, 3) # 2 x 3 \ud06c\uae30\uc758 \ub79c\ub364 Tensor\nprint(random_tensor)\n <\/code>\n <\/pre>\n\n \uc704 \ucf54\ub4dc\ub294 \ub2e4\uc591\ud55c \ud615\ud0dc\uc758 Tensor\ub97c \uc0dd\uc131\ud558\ub294 \uc608\uc81c\uc785\ub2c8\ub2e4.
\n Tensor\ub294 \ub2e4\uc591\ud55c \uc218\ud559\uc801 \uc5f0\uc0b0\uc5d0 \uc0ac\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n4.2 Tensor \uc5f0\uc0b0<\/h3>\n
\n Tensor\ub294 \ub2e4\uc591\ud55c \uc218\ud559\uc801 \uc5f0\uc0b0\uc744 \uc9c0\uc6d0\ud569\ub2c8\ub2e4. \uc5ec\uae30\uc11c\ub294 \uae30\ubcf8\uc801\uc778 \ud150\uc11c \uc5f0\uc0b0 \uba87 \uac00\uc9c0\ub97c \ub2e4\ub904\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \n# Tensor\uc758 \ud569\na = torch.tensor([1, 2])\nb = torch.tensor([3, 4])\nc = a + b\nprint(c)\n\n# \ud589\ub82c \uacf1\nx = torch.randn(2, 3)\ny = torch.randn(3, 2)\nz = torch.matmul(x, y)\nprint(z)\n\n# Tensor\uc758 \uc804\uce58\ntranspose = z.t()\nprint(transpose)\n <\/code>\n <\/pre>\n<\/section>\n\n4.3 Autograd<\/h3>\n
\n Autograd\ub294 \ud30c\uc774\ud1a0\uce58\uc758 \uc790\ub3d9 \ubbf8\ubd84 \uc2dc\uc2a4\ud15c\uc785\ub2c8\ub2e4.
\n \uc774\ub294 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc758 \ud559\uc2b5 \uacfc\uc815\uc5d0\uc11c \uae30\uc6b8\uae30\ub97c \uc790\ub3d9\uc73c\ub85c \uacc4\uc0b0\ud560 \uc218 \uc788\ub3c4\ub85d \ub3c4\uc640\uc90d\ub2c8\ub2e4. \uc608\uc81c \ucf54\ub4dc\ub97c \ud1b5\ud574 \uc54c\uc544\ubd05\uc2dc\ub2e4.\n <\/p>\n
\n \n# Tensor\uc5d0 requires_grad=True \uc124\uc815\nx = torch.tensor([[1.0, 2.0], [3.0, 4.0]], requires_grad=True)\ny = x ** 2 + 2 * x + 1 # y\ub97c x\uc5d0 \ub300\ud574 \uc815\uc758\n\n# y\uc5d0 \ub300\ud55c x\uc758 \uae30\uc6b8\uae30 \uacc4\uc0b0\ny.backward(torch.ones_like(y))\nprint(x.grad) # x\uc5d0 \ub300\ud55c \uae30\uc6b8\uae30 \ucd9c\ub825\n <\/code>\n <\/pre>\n<\/section>\n\n5. \uac04\ub2e8\ud55c \uc120\ud615 \ud68c\uadc0 \ubaa8\ub378 \uad6c\ud604\ud558\uae30<\/h2>\n
\n \uc774\uc81c \ud30c\uc774\ud1a0\uce58\ub97c \uc774\uc6a9\ud558\uc5ec \uc120\ud615 \ud68c\uadc0 \ubaa8\ub378\uc744 \uad6c\ud604\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.
\n \ud559\uc2b5 \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\uace0, \ubaa8\ub378\uc744 \uc815\uc758\ud55c \ud6c4 \ud559\uc2b5\uc744 \uc9c4\ud589\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \n# 1. \ub370\uc774\ud130 \uc0dd\uc131\nimport numpy as np\nimport torch\n\n# \uc784\uc758\uc758 \ub370\uc774\ud130 \uc0dd\uc131\nX = np.random.rand(100, 1).astype(np.float32) * 10 # 0\ubd80\ud130 10 \uc0ac\uc774\uc758 \uac12\ny = 2 * X + 1 + np.random.randn(100, 1).astype(np.float32) # y = 2x + 1 + noise\n\n# 2. Tensor\ub85c \ubcc0\ud658\nX_tensor = torch.from_numpy(X)\ny_tensor = torch.from_numpy(y)\n\n# 3. \ubaa8\ub378 \uc815\uc758\nmodel = torch.nn.Sequential(\n torch.nn.Linear(1, 1) # \uc785\ub825 1\uac1c, \ucd9c\ub825 1\uac1c\n)\n\n# 4. \uc190\uc2e4 \ud568\uc218 \ubc0f \uc635\ud2f0\ub9c8\uc774\uc800 \uc815\uc758\ncriterion = torch.nn.MSELoss()\noptimizer = torch.optim.SGD(model.parameters(), lr=0.01)\n\n# 5. \ubaa8\ub378 \ud559\uc2b5\nfor epoch in range(100):\n model.train()\n \n # \uc21c\uc804\ud30c\n y_pred = model(X_tensor)\n\n # \uc190\uc2e4 \uacc4\uc0b0\n loss = criterion(y_pred, y_tensor)\n\n # \uae30\uc6b8\uae30 \ucd08\uae30\ud654\n optimizer.zero_grad()\n\n # \uc5ed\uc804\ud30c\n loss.backward()\n\n # \uac00\uc911\uce58 \uc5c5\ub370\uc774\ud2b8\n optimizer.step()\n\n if (epoch+1) % 10 == 0:\n print(f'Epoch [{epoch+1}\/100], Loss: {loss.item():.4f}')\n <\/code>\n <\/pre>\n\n \uc704 \uc608\uc81c\ub294 \ub9e4\uc6b0 \uac04\ub2e8\ud55c \uc120\ud615 \ud68c\uadc0 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \uacfc\uc815\uc785\ub2c8\ub2e4.
\n \ub370\uc774\ud130 \uc0dd\uc131\ubd80\ud130 \ubaa8\ub378 \ud559\uc2b5\uae4c\uc9c0\uc758 \ubaa8\ub4e0 \uacfc\uc815\uc744 \ud3ec\ud568\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4.
\n \uacb0\uad6d \ud559\uc2b5\uc774 \uc644\ub8cc\ub41c \ud6c4, \ubaa8\ub378\uc758 \ud30c\ub77c\ubbf8\ud130\ub97c \ud655\uc778\ud574\ubcf4\uba74 2\uc640 1\uc5d0 \uac00\uae4c\uc6b4 \uac12\uc73c\ub85c \uc218\ub834\ud558\ub294 \uac83\uc744 \ubcfc \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n<\/section>\n\n6. \uacb0\ub860<\/h2>\n
\n \ud30c\uc774\ud1a0\uce58\ub294 \ub525\ub7ec\ub2dd\uc744 \uc704\ud55c \uac15\ub825\ud558\uace0 \uc720\uc5f0\ud55c \ub3c4\uad6c\uc785\ub2c8\ub2e4.
\n \uc774 \uae00\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\uc758 \uae30\ucd08 \ubb38\ubc95\uacfc \ud150\uc11c \uc0dd\uc131 \ubc0f \uc5f0\uc0b0, \uc790\ub3d9 \ubbf8\ubd84, \uac04\ub2e8\ud55c \uc120\ud615 \ud68c\uadc0 \ubaa8\ub378 \uad6c\ud604\uc744 \ub2e4\ub8e8\uc5c8\uc2b5\ub2c8\ub2e4.
\n \uc774\ub7ec\ud55c \uae30\ucd08\ub97c \ubc14\ud0d5\uc73c\ub85c \ud5a5\ud6c4 \ub354 \ubcf5\uc7a1\ud55c \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ucd95\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.
\n \uc55e\uc73c\ub85c \ud30c\uc774\ud1a0\uce58\ub85c \ub354 \ub9ce\uc740 \ud504\ub85c\uc81d\ud2b8\uc640 \uc5f0\uad6c\ub97c \uc9c4\ud589\ud574\ubcf4\uae38 \ubc14\ub78d\ub2c8\ub2e4.\n <\/p>\n<\/section>\n