\uc791\uc131\uc790: \uc870\uad11\ud615 | \ub0a0\uc9dc: [\ub0a0\uc9dc]<\/p>\n<\/header>\n \uc790\uc5f0\uc5b4 \ucc98\ub9ac(Natural Language Processing, NLP)\ub294 \ucef4\ud4e8\ud130\uac00 \uc778\uac04\uc758 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ubd84\uc11d\ud558\ub294 \uae30\uc220\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \ub525 \ub7ec\ub2dd\uc758 \ubc1c\uc804\uc73c\ub85c \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\ub294 \ube44\uc57d\uc801\uc778 \ubc1c\uc804\uc744 \uc774\ub8e8\uc5c8\uc73c\uba70, \uadf8 \uc911\uc5d0\uc11c\ub3c4 \uc21c\ud658 \uc2e0\uacbd\ub9dd(Recurrent Neural Network, RNN)\uc740 \uc5b8\uc5b4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \ub9e4\uc6b0 \ud6a8\uacfc\uc801\uc778 \ubaa8\ub378\ub85c \uc8fc\ubaa9\ubc1b\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ubc88 \uae00\uc5d0\uc11c\ub294 RNN\uc758 \uc6d0\ub9ac, \uad6c\uc870, \uadf8\ub9ac\uace0 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0\uc11c\uc758 \uc751\uc6a9\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \ubaa9\uc801\uc740 \ucef4\ud4e8\ud130\uac00 \uc778\uac04 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ud65c\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d \ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \uc8fc\uc694 \uacfc\uc81c\ub294 \uc5b8\uc5b4\uc758 \ubaa8\ud638\uc131\uc744 \ud574\uacb0\ud558\uace0, \ubb38\ub9e5\uc744 \uc774\ud574\ud558\uba70, \uc758\ubbf8\ub97c \ucd94\ub860\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uacfc\uc81c\ub4e4\uc744 \uc131\uacf5\uc801\uc73c\ub85c \uc218\ud589\ud558\uae30 \uc704\ud574 \ub2e4\uc591\ud55c \ubaa8\ub378\uc774 \uac1c\ubc1c\ub418\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n \uba38\uc2e0 \ub7ec\ub2dd\uc740 \ub370\uc774\ud130\ub97c \uc774\uc6a9\ud558\uc5ec \ud559\uc2b5\ud558\uace0 \uc608\uce21\ud558\ub294 \uc54c\uace0\ub9ac\uc998\uc744 \uc5f0\uad6c\ud558\ub294 \ubd84\uc57c\uc785\ub2c8\ub2e4. \ub525 \ub7ec\ub2dd\uc740 \uba38\uc2e0 \ub7ec\ub2dd\uc758 \ud558\uc704 \ubd84\uc57c\ub85c, \uc778\uacf5\uc2e0\uacbd\ub9dd\uc744 \uae30\ubc18\uc73c\ub85c \ubcf5\uc7a1\ud55c \uad6c\uc870\uc758 \ub370\uc774\ud130\uc5d0\uc11c \ud328\ud134\uc744 \ud559\uc2b5\ud558\ub294 \ubc29\ubc95\ub860\uc785\ub2c8\ub2e4. RNN\uc740 \uc774\ub7ec\ud55c \ub525 \ub7ec\ub2dd\uc758 \ud55c \uc885\ub958\ub85c, \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \ucd5c\uc801\ud654\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n RNN\uc740 \uc21c\ucc28\uc801\uc778 \ub370\uc774\ud130, \uc989 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\uae30 \uc704\ud574 \uc124\uacc4\ub41c \uc2e0\uacbd\ub9dd\uc785\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc778 \uc2e0\uacbd\ub9dd\uc740 \uc785\ub825 \ub370\uc774\ud130 \uac04\uc758 \uad00\uacc4\ub97c \ub3c5\ub9bd\uc801\uc73c\ub85c \ucc98\ub9ac\ud558\ub294 \ubc18\uba74, RNN\uc740 \uc774\uc804 \uc785\ub825\uc5d0 \ub300\ud55c \uc815\ubcf4\ub97c \uae30\uc5b5\ud558\uace0 \ud65c\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub294 \ud14d\uc2a4\ud2b8, \uc74c\uc131, \uc74c\uc545 \ub4f1\uacfc \uac19\uc740 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \ub9e4\uc6b0 \uc720\uc6a9\ud569\ub2c8\ub2e4.<\/p>\n<\/section>\n RNN\uc758 \uae30\ubcf8 \uad6c\uc870\ub294 \uc785\ub825\uce35, \uc740\ub2c9\uce35, \ucd9c\ub825\uce35\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \uc785\ub825\uce35\uc740 \ub2e8\uc5b4 \ub610\ub294 \ubb38\uc790 \uac19\uc740 \uc785\ub825 \ub370\uc774\ud130\ub97c \ubc1b\uc544\ub4e4\uc774\uace0, \uc740\ub2c9\uce35\uc740 \uc774\uc804\uc758 \uc0c1\ud0dc\ub97c \uae30\uc5b5\ud558\ub294 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. \ucd9c\ub825\uce35\uc740 \ucd5c\uc885\uc801\uc778 \uc608\uce21 \uacb0\uacfc\ub97c \uc81c\uacf5\ud569\ub2c8\ub2e4.<\/p>\n RNN\uc758 \uac00\uc7a5 \ud070 \ud2b9\uc9d5\uc740 \uc740\ub2c9 \uc0c1\ud0dc(hidden state)\uc785\ub2c8\ub2e4. \ub9e4 \uc2dc\uac04 t\uc5d0\uc11c\uc758 \uc740\ub2c9 \uc0c1\ud0dc\ub294 \uc774\uc804 \uc2dc\uac04 t-1\uc758 \uc740\ub2c9 \uc0c1\ud0dc\uc640 \ud604\uc7ac \uc785\ub825\uac12\uc744 \ubc14\ud0d5\uc73c\ub85c \uacc4\uc0b0\ub429\ub2c8\ub2e4. \uc774\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \uc218\uc2dd\uc73c\ub85c \ud45c\ud604\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n \uc5ec\uae30\uc11c RNN\uc740 \ub2e8\uae30 \uc758\uc874\uc131(short-term dependency) \ubb38\uc81c\ub97c \uc798 \ud574\uacb0\ud560 \uc218 \uc788\uc9c0\ub9cc \uc7a5\uae30 \uc758\uc874\uc131(long-term dependency) \ubb38\uc81c\uc5d0 \ucde8\uc57d\ud569\ub2c8\ub2e4. \uc774\ub294 RNN\uc774 \uc2dc\uac04\uc774 \uc9c0\ub0a8\uc5d0 \ub530\ub77c \uacfc\uac70\uc758 \uc815\ubcf4\ub97c \uc78a\uc5b4\ubc84\ub9ac\ub294 \uacbd\ud5a5\uc774 \uc788\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 LSTM(Long Short-Term Memory)\uc774\ub098 GRU(Gated Recurrent Unit)\uc640 \uac19\uc740 \ubcc0\ud615 \ubaa8\ub378\uc774 \uac1c\ubc1c\ub418\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n LSTM\uc740 RNN\uc758 \ubcc0\ud615\uc73c\ub85c, \uc7a5\uae30 \uc758\uc874\uc131 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 \ud2b9\ubcc4\ud55c \uba54\ubaa8\ub9ac \uc140(cell) \uad6c\uc870\ub97c \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. LSTM\uc758 \uc8fc\uc694 \uad6c\uc131 \uc694\uc18c\ub294 \uc785\ub825 \uac8c\uc774\ud2b8(input gate), \ub9dd\uac01 \uac8c\uc774\ud2b8(forget gate), \ucd9c\ub825 \uac8c\uc774\ud2b8(output gate)\ub85c \uc774\ub8e8\uc5b4\uc838 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \uad6c\uc870\ub97c \ud1b5\ud574 LSTM\uc740 \uc815\ubcf4\ub97c \uc120\ud0dd\uc801\uc73c\ub85c \uae30\uc5b5\ud558\uace0 \uc78a\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n GRU\ub294 LSTM\uacfc \uc720\uc0ac\ud558\uc9c0\ub9cc \ub354 \uac04\ub2e8\ud55c \uad6c\uc870\ub97c \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. GRU\ub294 \uc5c5\ub370\uc774\ud2b8 \uac8c\uc774\ud2b8(update gate)\uc640 \ub9ac\uc14b \uac8c\uc774\ud2b8(reset gate)\ub97c \ud1b5\ud574 \uc815\ubcf4\uc758 \ud750\ub984\uc744 \uc870\uc808\ud569\ub2c8\ub2e4. GRU\ub294 \uc77c\ubc18\uc801\uc73c\ub85c LSTM\ubcf4\ub2e4 \uacc4\uc0b0\uc774 \ub35c \ubcf5\uc7a1\ud558\uc5ec \ub354 \ube60\ub974\uac8c \ud559\uc2b5\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n RNN\uc740 \uae30\uacc4 \ubc88\uc5ed \ubd84\uc57c\uc5d0\uc11c \ub9e4\uc6b0 \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc785\ub825 \ubb38\uc7a5\uc744 RNN\uc744 \ud1b5\ud574 \uc778\ucf54\ub529\ud55c \ud6c4, \ucd9c\ub825 \ubb38\uc7a5\uc744 \uc0dd\uc131\ud558\ub294 \ub514\ucf54\ub354 \uc5ed\ud560\uc744 \uc218\ud589\ud569\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \uc77c\ubc18\uc801\uc73c\ub85c Encoder-Decoder \uad6c\uc870\ub85c \uad6c\ud604\ub429\ub2c8\ub2e4.<\/p>\n RNN\uc740 \ud14d\uc2a4\ud2b8\uc758 \uac10\uc815\uc744 \ubd84\uc11d\ud558\ub294 \ub370\ub3c4 \ub110\ub9ac \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uc758 \uc2dc\ud000\uc2a4\ub97c \uc785\ub825\uc73c\ub85c \ubc1b\uace0, \ud574\ub2f9 \ud14d\uc2a4\ud2b8\uc758 \uac10\uc815\uc744 \ud310\ub2e8\ud558\uae30 \uc704\ud574 \uac01 \uc2dc\uc810\uc5d0\uc11c \uc740\ub2c9 \uc0c1\ud0dc\uac00 \uc5c5\ub370\uc774\ud2b8\ub429\ub2c8\ub2e4.<\/p>\n RNN\uc744 \ud65c\uc6a9\ud558\uc5ec \ud14d\uc2a4\ud2b8 \uc0dd\uc131 \ubaa8\ub378\uc744 \ub9cc\ub4e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc8fc\uc5b4\uc9c4 \ub2e8\uc5b4 \uc2dc\ud000\uc2a4\ub97c \ubc14\ud0d5\uc73c\ub85c \ub2e4\uc74c\uc5d0 \uc62c \ub2e8\uc5b4\ub97c \uc608\uce21\ud568\uc73c\ub85c\uc368, \uc790\uc5f0\uc2a4\ub7ec\uc6b4 \ubb38\uc7a5\uc744 \uc0dd\uc131\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n<\/section>\n \uc544\ub798\ub294 Python\uacfc TensorFlow\ub97c \uc774\uc6a9\ud55c \uac04\ub2e8\ud55c RNN \ubaa8\ub378\uc758 \uc608\uc81c\uc785\ub2c8\ub2e4.<\/p>\n \uc774\ubc88 \uae00\uc5d0\uc11c\ub294 RNN\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \uc791\ub3d9 \uc6d0\ub9ac, \uadf8\ub9ac\uace0 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0\uc11c\uc758 \ud65c\uc6a9 \uc0ac\ub840\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. RNN\uc740 \uc5ec\uc804\ud788 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud558\uace0 \uc788\uc73c\uba70, LSTM\uacfc GRU\uc640 \uac19\uc740 \ubcc0\ud615 \ubaa8\ub378\uc744 \ud1b5\ud574 \uc7a5\uae30 \uc758\uc874\uc131 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \ub525 \ub7ec\ub2dd\uc758 \ubc1c\uc804\uc5d0 \ub530\ub77c \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\uc220\uc740 \uacc4\uc18d\ud574\uc11c \ubc1c\uc804\ud560 \uac83\uc73c\ub85c \uae30\ub300\ub429\ub2c8\ub2e4.<\/p>\n<\/section>\n1. \uc11c\ub860<\/h2>\n
2. \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uac1c\uc694<\/h2>\n
3. \uba38\uc2e0 \ub7ec\ub2dd\uacfc \ub525 \ub7ec\ub2dd\uc758 \uad00\uacc4<\/h2>\n
4. \uc21c\ud658 \uc2e0\uacbd\ub9dd(Recurrent Neural Network, RNN) \uac1c\ub150<\/h2>\n
5. RNN\uc758 \uad6c\uc870\uc640 \uc791\ub3d9 \uc6d0\ub9ac<\/h2>\n
5.1. \uae30\ubcf8 \uad6c\uc870<\/h3>\n
5.2. \uc0c1\ud0dc \uc804\ud30c<\/h3>\n
<\/p>\n
h_t<\/code>\ub294 \ud604\uc7ac \uc2dc\uc810\uc758 \uc740\ub2c9 \uc0c1\ud0dc, f<\/code>\ub294 \ud65c\uc131\ud654 \ud568\uc218, W_hh<\/code>\ub294 \uc740\ub2c9 \uc0c1\ud0dc \uac04\uc758 \uac00\uc911\uce58, W_xh<\/code>\ub294 \uc785\ub825\uacfc \uc740\ub2c9 \uc0c1\ud0dc \uac04\uc758 \uac00\uc911\uce58\uc785\ub2c8\ub2e4.<\/p>\n<\/section>\n6. RNN\uc758 \ud55c\uacc4<\/h2>\n
7. LSTM\uacfc GRU<\/h2>\n
7.1. LSTM<\/h3>\n
7.2. GRU<\/h3>\n
8. \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0\uc11c RNN\uc758 \ud65c\uc6a9<\/h2>\n
8.1. \uae30\uacc4 \ubc88\uc5ed<\/h3>\n
8.2. \uac10\uc815 \ubd84\uc11d<\/h3>\n
8.3. \ud14d\uc2a4\ud2b8 \uc0dd\uc131<\/h3>\n
9. RNN\uc758 \uc2e4\uc81c \uad6c\ud604 \uc608\uc81c<\/h2>\n
\nimport tensorflow as tf\nfrom tensorflow.keras import layers\n\n# \ub370\uc774\ud130 \uc900\ube44\n# (\uc5ec\uae30\uc11c\ub294 \ub370\uc774\ud130 \ub85c\ub529 \ubc0f \uc804\ucc98\ub9ac \ucf54\ub4dc\uac00 \uc0dd\ub7b5\ub428)\n\n# \ubaa8\ub378 \uc815\uc758\nmodel = tf.keras.Sequential()\nmodel.add(layers.SimpleRNN(128, input_shape=(None, number_of_features)))\nmodel.add(layers.Dense(number_of_classes, activation='softmax'))\n\n# \ubaa8\ub378 \ucef4\ud30c\uc77c\nmodel.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n\n# \ubaa8\ub378 \ud559\uc2b5\nmodel.fit(X_train, y_train, epochs=10, batch_size=32)\n <\/code><\/pre>\n<\/section>\n10. \uacb0\ub860<\/h2>\n