\ucd5c\uadfc \uba87 \ub144 \uac04 \uc778\uacf5\uc9c0\ub2a5\uacfc \uba38\uc2e0\ub7ec\ub2dd \uae30\uc220\uc774 \ub180\ub77c\uc6b4 \ubc1c\uc804\uc744 \uc774\ub8e8\uba70, \uae08\uc735 \uc2dc\uc7a5 \ud2b9\ud788 \uc54c\uace0\ub9ac\uc998 \ud2b8\ub808\uc774\ub529 \ubd84\uc57c\uc5d0\uc11c\ub3c4 \uadf8 \uc911\uc694\uc131\uc774 \ucee4\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \uac15\uc88c\uc5d0\uc11c\ub294 \ub525\ub7ec\ub2dd\uc758 \ud55c \uac00\uc9c0 \ubaa8\ub378\uc778 \uc0dd\uc131\uc801 \uc801\ub300 \uc2e0\uacbd\ub9dd(Generative Adversarial Network, GAN)\uc744 \uc0ac\uc6a9\ud558\uc5ec \ub370\uc774\ud130 \uc0dd\uc131 \ubc0f \uc774 \ub370\uc774\ud130\ub97c \uae30\ubc18\uc73c\ub85c \ud55c \ud2b8\ub808\uc774\ub529 \uc804\ub7b5 \uc218\ub9bd \ubc29\ubc95\uc5d0 \ub300\ud574 \ub2e4\ub8f0 \uac83\uc785\ub2c8\ub2e4. \ud2b9\ud788 TensorFlow 2\ub97c \ud65c\uc6a9\ud558\uc5ec GAN\uc744 \uad6c\ucd95\ud558\ub294 \ubc29\ubc95\uc744 \ub2e8\uacc4\ubcc4\ub85c \uc548\ub0b4\ud569\ub2c8\ub2e4.<\/p>\n
GAN\uc740 \ub450 \uac1c\uc758 \uc2e0\uacbd\ub9dd, \uc989 \uc0dd\uc131\uc790(Generator)\uc640 \ud310\ubcc4\uc790(Discriminator)\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \uc0dd\uc131\uc790\ub294 \uc2e4\uc81c \ub370\uc774\ud130\ucc98\ub7fc \ubcf4\uc774\ub294 \uac00\uc9dc \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\uace0, \ud310\ubcc4\uc790\ub294 \uc774 \ub370\uc774\ud130\uac00 \uc2e4\uc81c \ub370\uc774\ud130\uc778\uc9c0 \uc0dd\uc131\uc790\uac00 \ub9cc\ub4e0 \ub370\uc774\ud130\uc778\uc9c0\ub97c \ud310\ub2e8\ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ub450 \uc2e0\uacbd\ub9dd\uc740 \uc11c\ub85c \uacbd\uc7c1\ud558\uba70 \ud559\uc2b5\ud558\uac8c \ub418\ub294\ub370, \uc774\ub97c \ud1b5\ud574 \uc0dd\uc131\uc790\ub294 \ub354\uc6b1 \uc2e4\uc81c\uc640 \uc720\uc0ac\ud55c \ub370\uc774\ud130\ub97c \ub9cc\ub4e4\uc5b4\ub0bc \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
GAN\uc758 \uae30\ubcf8 \uad6c\uc870\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
GAN\uc758 \ud559\uc2b5 \uacfc\uc815\uc740 \ub300\ub7b5\uc801\uc73c\ub85c \ub2e4\uc74c\uacfc \uac19\uc740 \ub2e8\uacc4\ub97c \ud3ec\ud568\ud569\ub2c8\ub2e4:<\/p>\n
\uc774\uc81c TensorFlow 2\ub97c \uc0ac\uc6a9\ud558\uc5ec GAN\uc744 \uad6c\ud604\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c\ub294 \uae30\ubcf8\uc801\uc778 GAN\uc758 \uad6c\uc131 \uc694\uc18c\ub97c \uc124\uba85\ud558\uace0, \uc774\ub97c \uae08\uc735 \ub370\uc774\ud130\uc5d0 \uc801\uc6a9\ud558\ub294 \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
TensorFlow 2\uc640 \uae30\ud0c0 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud569\ub2c8\ub2e4. \ub2e4\uc74c \uba85\ub839\uc5b4\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
pip install tensorflow numpy matplotlib<\/code><\/pre>\n3.2. \ub370\uc774\ud130 \ubd88\ub7ec\uc624\uae30<\/h3>\n
\uc8fc\uc2dd \uc2dc\uc7a5 \ub370\uc774\ud130\ub97c \uc5bb\uae30 \uc704\ud574 Yahoo Finance API\uc640 \uac19\uc740 \uacf5\uac1c API\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\ub294 \ub370\uc774\ud130 \ub85c\ub4dc \ubc0f \uc804\ucc98\ub9ac \ubc29\ubc95\uc785\ub2c8\ub2e4.<\/p>\n
import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport yfinance as yf\n\n# \uc8fc\uc2dd \ub370\uc774\ud130 \ub85c\ub4dc\ndata = yf.download(\"AAPL\", start=\"2010-01-01\", end=\"2020-01-01\")\ndata = data['Close'].values.reshape(-1, 1)\ndata = (data - np.mean(data)) \/ np.std(data) # \uc815\uaddc\ud654<\/code><\/pre>\n3.3. GAN \ubaa8\ub378 \uad6c\ucd95<\/h3>\n
\uc774\uc81c GAN\uc758 \uc0dd\uc131\uc790\uc640 \ud310\ubcc4\uc790\ub97c \uad6c\ucd95\ud560 \ucc28\ub840\uc785\ub2c8\ub2e4. TensorFlow Keras API\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac04\ub2e8\ud55c \ubaa8\ub378\uc744 \uad6c\ud604\ud574\ubd05\uc2dc\ub2e4.<\/p>\n
import tensorflow as tf\nfrom tensorflow.keras import layers\n\n# \uc0dd\uc131\uc790 \ubaa8\ub378\ndef build_generator():\n model = tf.keras.Sequential()\n model.add(layers.Dense(128, activation='relu', input_shape=(100,)))\n model.add(layers.Dense(256, activation='relu'))\n model.add(layers.Dense(512, activation='relu'))\n model.add(layers.Dense(1, activation='tanh')) # \uc8fc\uc2dd \ub370\uc774\ud130\ub294 -1\uc5d0\uc11c 1\uae4c\uc9c0 \uc815\uaddc\ud654\n return model\n\n# \ud310\ubcc4\uc790 \ubaa8\ub378\ndef build_discriminator():\n model = tf.keras.Sequential()\n model.add(layers.Dense(512, activation='relu', input_shape=(1,)))\n model.add(layers.Dense(256, activation='relu'))\n model.add(layers.Dense(1, activation='sigmoid'))\n return model\n\ngenerator = build_generator()\ndiscriminator = build_discriminator()<\/code><\/pre>\n3.4. GAN \ud559\uc2b5<\/h3>\n
GAN\uc744 \ud559\uc2b5\ud558\uae30 \uc704\ud574 \ud544\uc694\ud55c \uc190\uc2e4 \ud568\uc218\uc640 \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc124\uc815\ud558\uace0, \ud559\uc2b5 \ub8e8\ud504\ub97c \uad6c\uc131\ud569\ub2c8\ub2e4.<\/p>\n
loss_fn = tf.keras.losses.BinaryCrossentropy()\ngenerator_optimizer = tf.keras.optimizers.Adam(1e-4)\ndiscriminator_optimizer = tf.keras.optimizers.Adam(1e-4)\n\n# GAN \ud559\uc2b5 \ub8e8\ud504\ndef train_gan(epochs, batch_size):\n for epoch in range(epochs):\n for _ in range(batch_size):\n noise = np.random.normal(0, 1, size=(batch_size, 100))\n generated_data = generator(noise)\n\n idx = np.random.randint(0, data.shape[0], batch_size)\n real_data = data[idx]\n\n with tf.GradientTape() as disc_tape:\n real_output = discriminator(real_data)\n fake_output = discriminator(generated_data)\n\n disc_loss = loss_fn(tf.ones_like(real_output), real_output) + \\\n loss_fn(tf.zeros_like(fake_output), fake_output)\n\n gradients = disc_tape.gradient(disc_loss, discriminator.trainable_variables)\n discriminator_optimizer.apply_gradients(zip(gradients, discriminator.trainable_variables))\n\n with tf.GradientTape() as gen_tape:\n fake_output = discriminator(generated_data)\n gen_loss = loss_fn(tf.ones_like(fake_output), fake_output)\n\n gradients = gen_tape.gradient(gen_loss, generator.trainable_variables)\n generator_optimizer.apply_gradients(zip(gradients, generator.trainable_variables))\n\n if epoch % 100 == 0:\n print(f'Epoch {epoch} - Discriminator Loss: {disc_loss.numpy()} - Generator Loss: {gen_loss.numpy()}')\n\ntrain_gan(epochs=10000, batch_size=32)<\/code><\/pre>\n4. GAN \uacb0\uacfc \ubd84\uc11d<\/h2>\n
\ud559\uc2b5\uc774 \uc644\ub8cc\ub41c \ud6c4, \uc0dd\uc131\ub41c \ub370\uc774\ud130\uac00 \uc2e4\uc81c \ub370\uc774\ud130\uc640 \uc5bc\ub9c8\ub098 \uc720\uc0ac\ud55c\uc9c0\ub97c \uc2dc\uac01\ud654\ud558\uc5ec \ubd84\uc11d\ud574\ubd05\ub2c8\ub2e4.<\/p>\n
def plot_generated_data(generator, num_samples=1000):\n noise = np.random.normal(0, 1, size=(num_samples, 100))\n generated_data = generator(noise)\n\n plt.figure(figsize=(10, 5))\n plt.plot(generated_data, label='Generated Data')\n plt.plot(data[0:num_samples], label='Real Data')\n plt.legend()\n plt.show()\n\nplot_generated_data(generator)<\/code><\/pre>\n5. \uacb0\ub860<\/h2>\n
\ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \uba38\uc2e0\ub7ec\ub2dd \ubc0f \ub525\ub7ec\ub2dd \uae30\ubc18\uc758 \uc0dd\uc131\uc801 \uacbd\uc7c1 \uc2e0\uacbd\ub9dd\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc8fc\uc2dd \uc2dc\uc7a5 \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\uace0, \uc774 \ub370\uc774\ud130\ub97c \ud1b5\ud574 \uc7a0\uc7ac\uc801\uc778 \ud2b8\ub808\uc774\ub529 \uc804\ub7b5\uc744 \uac1c\ubc1c\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. GAN\uc740 \ub2e4\uc591\ud55c \ub370\uc774\ud130 \uc0dd\uc131 \uc791\uc5c5\uc5d0 \ud6a8\uacfc\uc801\uc774\uba70, \uc54c\uace0\ub9ac\uc998 \ud2b8\ub808\uc774\ub529\uc5d0\uc11c\ub3c4 \ub9e4\uc6b0 \uc720\uc6a9\ud558\uac8c \ud65c\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud5a5\ud6c4 \ub354 \ubc1c\uc804\ub41c \ubaa8\ub378\uacfc \uae30\ubc95\uc744 \ud1b5\ud574 \uc774 \ubd84\uc57c\uc5d0\uc11c\uc758 \uac00\ub2a5\uc131\uc744 \ub354\uc6b1 \ud0d0\uad6c\ud574\ubcf4\ub294 \uac83\uc744 \ucd94\ucc9c\ud569\ub2c8\ub2e4.<\/p>\n
6. \ucc38\uace0 \uc790\ub8cc<\/h2>\n