\ub525 \ub7ec\ub2dd(Deep Learning)\uc740 \uba38\uc2e0 \ub7ec\ub2dd\uc758 \ud55c \uc885\ub958\ub85c, Layers\uc758 \uc9d1\ud569\uc778 \uc2e0\uacbd\ub9dd(Neural Network)\uc744 \uc0ac\uc6a9\ud558\uc5ec \ub370\uc774\ud130\uc758 \ud2b9\uc9d5\uc744 \ud559\uc2b5\ud569\ub2c8\ub2e4. \uc790\uc5f0\uc5b4 \ucc98\ub9ac(Natural Language Processing, NLP)\ub294 \ucef4\ud4e8\ud130\uac00 \uc790\uc5f0\uc5b4\ub97c \uc774\ud574\ud558\uace0 \uc0dd\uc131\ud558\ub294 \uae30\uc220\ub85c, \ud14d\uc2a4\ud2b8 \ubd84\uc11d, \ubc88\uc5ed, \uc74c\uc131 \uc778\uc2dd \ub4f1 \ub2e4\uc591\ud55c \uc751\uc6a9 \ubd84\uc57c\uac00 \uc788\uc2b5\ub2c8\ub2e4. \ud2b9\ud788 \ud0dc\uae45(Tagging) \uc791\uc5c5\uc740 \uac01 \ub2e8\uc5b4\uc5d0 \ud45c\uc2dd\uc744 \ubd80\uc5ec\ud558\ub294 \uacfc\uc815\uc73c\ub85c, \uc608\ub97c \ub4e4\uc5b4 \ud488\uc0ac \ud0dc\uae45(Part-of-Speech Tagging)\uc5d0\uc11c\ub294 \uac01 \ub2e8\uc5b4\uc5d0 \uba85\uc0ac, \ub3d9\uc0ac, \ud615\uc6a9\uc0ac \ub4f1\uc758 \ud0dc\uadf8\ub97c \ud560\ub2f9\ud569\ub2c8\ub2e4.<\/p>\n
\uc790\uc5f0\uc5b4 \ucc98\ub9ac\ub294 \ucd94\uc0c1\uc801\uc774\uace0 \ube44\uc815\ud615\ud654\ub41c \uc5b8\uc5b4 \ub370\uc774\ud130\ub97c \uad6c\uc870\ud654\ud558\ub294 \uc791\uc5c5\uc774 \ud3ec\ud568\ub429\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c \ud14d\uc2a4\ud2b8\ub97c \ubd84\ud574\ud558\uace0, \uc758\ubbf8\ub97c \ucd94\ucd9c\ud558\uba70, \ubb38\ub9e5\uc744 \uc774\ud574\ud558\ub294 \uac83\uc774 \uc911\uc694\ud569\ub2c8\ub2e4. \uc804\ud1b5\uc801\uc778 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\ubc95\uc778 \ud1b5\uacc4\uc801 \ubaa8\ub378\ub9c1\uc5d0\uc11c \ubc97\uc5b4\ub098, \ucd5c\uadfc\uc758 \ub525 \ub7ec\ub2dd \uae30\ubc18 \ubc29\ubc95\uc774 \ub192\uc740 \uc131\ub2a5\uc744 \ubcf4\uc774\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub525 \ub7ec\ub2dd \ubaa8\ub378\uc740 \uc5ec\ub7ec \uce35\uc73c\ub85c \uad6c\uc131\ub41c \uc778\uacf5 \uc2e0\uacbd\ub9dd\uc744 \uae30\ubc18\uc73c\ub85c \ud569\ub2c8\ub2e4. \uac01 \uce35\uc740 \ud2b9\uc815\ud55c \ub370\uc774\ud130 \ud45c\ud604\uc744 \ud559\uc2b5\ud558\uba70, \uc815\ubcf4\uac00 \uce35\uc744 \uac70\uce58\uba74\uc11c \uc810\uc810 \ub354 \ucd94\uc0c1\ud654\ub41c \ud615\ud0dc\ub85c \ubcc0\ud658\ub429\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubc29\uc2dd\uc740 \ud2b9\ud788 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0 \ud6a8\uacfc\uc801\uc774\uba70, \ubb38\ub9e5\uacfc \uc758\ubbf8\ub97c \uace0\ub824\ud55c \uace0\uae09 \ud45c\ud604\uc744 \ud559\uc2b5\ud558\ub294 \ub370 \uac15\uc810\uc744 \uac00\uc9d1\ub2c8\ub2e4.<\/p>\n
\ub525 \ub7ec\ub2dd \ubaa8\ub378\uc740 \uc785\ub825\uce35, \uc740\ub2c9\uce35, \ucd9c\ub825\uce35\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \uc785\ub825\uce35\uc740 \uc6d0\ub798 \ub370\uc774\ud130(\uc785\ub825 \ubca1\ud130)\ub97c \ubc1b\uace0, \uc740\ub2c9\uce35\uc740 \uc785\ub825\uc5d0\uc11c \ubcf5\uc7a1\ud55c \ud328\ud134\uc744 \ud559\uc2b5\ud558\uba70, \ucd9c\ub825\uce35\uc740 \ucd5c\uc885 \uacb0\uacfc(\uc608: \ubd84\ub958, \ud68c\uadc0)\ub97c \ucd9c\ub825\ud569\ub2c8\ub2e4.<\/p>\n
\n model = Sequential()\n model.add(Dense(128, input_dim=input_dim, activation='relu'))\n model.add(Dropout(0.5))\n model.add(Dense(num_classes, activation='softmax'))\n <\/code><\/pre>\n2.2 \ucf00\ub77c\uc2a4(Keras) \uc18c\uac1c<\/h3>\n
\ucf00\ub77c\uc2a4\ub294 \ud30c\uc774\uc36c\uc73c\ub85c \uc791\uc131\ub41c \uace0\uc218\uc900\uc758 \uc2e0\uacbd\ub9dd API\ub85c, TensorFlow, CNTK, Theano\uc640 \uac19\uc740 \uc800\uc218\uc900 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ubc31\uc5d4\ub4dc\ub85c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc9c1\uad00\uc801\uc778 \uc778\ud130\ud398\uc774\uc2a4\ub97c \uc81c\uacf5\ud558\uc5ec \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc27d\uac8c \uad6c\ucd95\ud558\uace0 \ud559\uc2b5\ud560 \uc218 \uc788\ub3c4\ub85d \ub3c4\uc640\uc90d\ub2c8\ub2e4.<\/p>\n
3. \ud0dc\uae45 \uc791\uc5c5\uc758 \uc815\uc758\uc640 \ud544\uc694\uc131<\/h2>\n
\ud0dc\uae45 \uc791\uc5c5\uc740 \uc8fc\uc5b4\uc9c4 \ud14d\uc2a4\ud2b8\uc758 \uac01 \ub2e8\uc5b4\uc5d0 \ud2b9\uc815\ud55c \uc815\ubcf4\ub97c \ubd80\uc5ec\ud558\ub294 \uacfc\uc815\uc73c\ub85c, \ubb38\ub9e5\uc744 \uc774\ud574\ud558\uace0 \ub2e4\uc591\ud55c \uc815\ubcf4 \ucc98\ub9ac\ub97c \uac00\ub2a5\ud558\uac8c \ud569\ub2c8\ub2e4. \ud488\uc0ac \ud0dc\uae45, \uac1c\uccb4\uba85 \uc778\uc2dd \ub4f1\uc73c\ub85c \ud655\uc7a5\ub420 \uc218 \uc788\ub294 \ud0dc\uae45 \uc791\uc5c5\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \ub9c8\uc9c0\ub9c9 \ub2e8\uacc4\uc5d0\uc11c\ub3c4 \uae30\ubcf8\uc801\uc778 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4.<\/p>\n
3.1 \ud0dc\uae45\uc758 \uc720\ud615<\/h3>\n\n- \ud488\uc0ac \ud0dc\uae45:<\/strong> \uac01 \ub2e8\uc5b4\uc5d0 \uba85\uc0ac, \ub3d9\uc0ac \ub4f1 \ud488\uc0ac \uc815\ubcf4\ub97c \ubd80\uc5ec\ud569\ub2c8\ub2e4.<\/li>\n
- \uac1c\uccb4\uba85 \uc778\uc2dd:<\/strong> \uc0ac\ub78c, \uc7a5\uc18c, \uae30\uad00 \ub4f1\uc744 \uc2dd\ubcc4\ud558\uc5ec \ud0dc\uae45\ud569\ub2c8\ub2e4.<\/li>\n
- \uc815\uc11c \ubd84\uc11d:<\/strong> \ud14d\uc2a4\ud2b8\uc758 \uac10\uc815\uc744 \ubd84\uc11d\ud558\uc5ec \uae0d\uc815, \ubd80\uc815 \ub4f1\uc758 \ud0dc\uadf8\ub97c \ubd80\uc5ec\ud569\ub2c8\ub2e4.<\/li>\n<\/ul>\n
4. Keras\ub97c \uc774\uc6a9\ud55c \ud0dc\uae45 \uc791\uc5c5 \uad6c\ud604<\/h2>\n
\ucf00\ub77c\uc2a4\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud0dc\uae45 \uc791\uc5c5\uc744 \uc9c4\ud589\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud55c \uad6c\uccb4\uc801\uc778 \uc808\ucc28\ub97c \ub2e4\ub904\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \ub370\uc774\ud130 \uc804\ucc98\ub9ac, \ubaa8\ub378 \uc815\uc758, \ud559\uc2b5, \ud3c9\uac00 \ub4f1\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4.<\/p>\n
4.1 \ub370\uc774\ud130 \uc804\ucc98\ub9ac<\/h3>\n
\uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0\uc11c \uac00\uc7a5 \uccab \ub2e8\uacc4\ub294 \ub370\uc774\ud130 \uc804\ucc98\ub9ac\uc785\ub2c8\ub2e4. \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\uc5ec \ubaa8\ub378\uc5d0 \uc801\ud569\ud55c \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658\ud574\uc57c \ud569\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\ub294 \ud1a0\ud070\ud654, \uc815\uc218 \uc778\ucf54\ub529, \ud328\ub529 \ub4f1\uc774 \ud3ec\ud568\ub429\ub2c8\ub2e4.<\/p>\n
\n from keras.preprocessing.text import Tokenizer\n from keras.preprocessing.sequence import pad_sequences\n\n tokenizer = Tokenizer()\n tokenizer.fit_on_texts(sentences)\n sequences = tokenizer.texts_to_sequences(sentences)\n padded_sequences = pad_sequences(sequences, maxlen=maxlen)\n <\/code><\/pre>\n4.2 \ubaa8\ub378 \uc815\uc758<\/h3>\n
\ub370\uc774\ud130\ub97c \uc804\ucc98\ub9ac\ud55c \ud6c4, Keras\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud0dc\uae45 \ubaa8\ub378\uc744 \uc815\uc758\ud569\ub2c8\ub2e4. LSTM(Long Short-Term Memory)\uc774\ub098 GRU(Gated Recurrent Unit)\uc640 \uac19\uc740 \uc21c\ud658 \uc2e0\uacbd\ub9dd\uc744 \ud0c0\uae43\uc73c\ub85c \uad6c\uc131\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\n from keras.models import Sequential\n from keras.layers import Embedding, LSTM, Dense, TimeDistributed\n\n model = Sequential()\n model.add(Embedding(input_dim=num_words, output_dim=embedding_dim, input_length=maxlen))\n model.add(LSTM(128, return_sequences=True))\n model.add(TimeDistributed(Dense(num_classes, activation='softmax')))\n <\/code><\/pre>\n4.3 \ubaa8\ub378 \ud559\uc2b5<\/h3>\n
\ubaa8\ub378 \uad6c\uc870\ub97c \uc815\uc758\ud55c \ud6c4, \uc190\uc2e4 \ud568\uc218\uc640 \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc124\uc815\ud558\uace0 \ud559\uc2b5\uc744 \uc9c4\ud589\ud569\ub2c8\ub2e4. \ubcf4\ud1b5, \ud06c\ub85c\uc2a4\uc5d4\ud2b8\ub85c\ud53c \uc190\uc2e4 \ud568\uc218\uc640 Adam \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc0ac\uc6a9\ud569\ub2c8\ub2e4.<\/p>\n
\n model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n model.fit(padded_sequences, labels, epochs=5, batch_size=32)\n <\/code><\/pre>\n4.4 \ubaa8\ub378 \ud3c9\uac00 \ubc0f \uc608\uce21<\/h3>\n
\ud6c8\ub828\uc774 \uc644\ub8cc\ub41c \ud6c4, \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4. \uc815\ud655\ub3c4(accuracy)\uc640 \uac19\uc740 \uc9c0\ud45c\ub97c \ud1b5\ud574 \ubaa8\ub378 \uc131\ub2a5\uc744 \ud310\ub2e8\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\n test_loss, test_acc = model.evaluate(test_sequences, test_labels)\n predictions = model.predict(new_sequences)\n <\/code><\/pre>\n5. \uacb0\ub860<\/h2>\n
\ub525 \ub7ec\ub2dd\uc744 \uc774\uc6a9\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\uc220\uc740 \ub0a0\uc774 \uac08\uc218\ub85d \uc131\uc7a5\ud558\uace0 \uc788\uc73c\uba70, \ucf00\ub77c\uc2a4\ub97c \ud1b5\ud574 \uc2e4\uc9c8\uc801\uc778 \ud0dc\uae45 \uc791\uc5c5\uc744 \uc190\uc27d\uac8c \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \ub354\uc6b1 \ub2e4\uc591\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\uc220\uc774 \ubc1c\uc804\ud558\uc5ec \uc6b0\ub9ac \uc0dd\ud65c \uc18d\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud560 \uac83\uc785\ub2c8\ub2e4. \ud0dc\uae45 \uc791\uc5c5\uc740 \uc774\ub7ec\ud55c \uae30\uc220\uc758 \uae30\ubcf8\uc774 \ub418\uc5b4, \ub354 \ub098\uc544\uac00 \ubcf5\uc7a1\ud55c \uc5b8\uc5b4 \uc774\ud574 \uc791\uc5c5\uc73c\ub85c \ud655\uc7a5\ub420 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\ucd94\uac00\uc801\uc73c\ub85c, \uae30\uacc4 \ud559\uc2b5\uacfc \ub525 \ub7ec\ub2dd\uc758 \ubc1c\uc804\uacfc \ud568\uaed8 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \uc815\ud655\ub3c4\uc640 \ud6a8\uc728\uc131\uc774 \ub192\uc544\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ud5a5\ud6c4 \uc5f0\uad6c\uc640 \uac1c\ubc1c\uc774 \uae30\ub300\ub418\ub294 \ubd84\uc57c\ub85c, \ub9ce\uc740 \ub370\uc774\ud130\uc640 \ub354 \ubc1c\uc804\ub41c \uc54c\uace0\ub9ac\uc998\uc774 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \ud488\uc9c8\uc744 \ub354\uc6b1 \ud5a5\uc0c1\uc2dc\ud0ac \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\uc774 \uae00\uc774 \ub525 \ub7ec\ub2dd\uc744 \ud65c\uc6a9\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac, \ud2b9\ud788 \ud0dc\uae45 \uc791\uc5c5\uc5d0 \ub300\ud55c \uc804\ubc18\uc801\uc778 \uc774\ud574\uc5d0 \ub3c4\uc6c0\uc774 \ub418\uc5c8\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4. \uac01 \uc8fc\uc81c\uc5d0 \ub300\ud574 \ub354 \uae4a\uc774 \uc54c\uace0 \uc2f6\uc73c\uc2e0 \ubd84\ub4e4\uc740 \uad00\ub828 \uc790\ub8cc\ub97c \ucc3e\uc544\ubcf4\uc2dc\uae30\ub97c \ucd94\ucc9c\ub4dc\ub9bd\ub2c8\ub2e4.<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"