\u3053\u3093\u306b\u3061\u306f\u3002python\u3092\u4f7f\u3063\u305f\u6a5f\u68b0\u5b66\u7fd2\u306b\u8208\u5473\u304c\u3042\u308b\u3051\u308c\u3069\u3001\u306a\u304b\u306a\u304b\u4e00\u6b69\u304c\u8e0f\u307f\u51fa\u305b\u306a\u3044\u3068\u3044\u3046\u65b9\u306f\u3044\u307e\u305b\u3093\u304b\uff1f\u305d\u3093\u306a\u3042\u306a\u305f\u306b\u304a\u3059\u3059\u3081\u3057\u305f\u3044\u306e\u304c\u3001Keras\u3067\u3059\u3002Keras\u306f\u3001\u30b7\u30f3\u30d7\u30eb\u3067\u4f7f\u3044\u3084\u3059\u304f\u3001\u304b\u3064\u5f37\u529b\u306a\u6a5f\u68b0\u5b66\u7fd2\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002\u672c\u8a18\u4e8b\u3067\u306f\u3001Keras\u306e\u57fa\u790e\u304b\u3089\u5b9f\u8df5\u7684\u306a\u30c6\u30af\u30cb\u30c3\u30af\u307e\u3067\u3092\u51dd\u7e2e\u3057\u3066\u304a\u4f1d\u3048\u3057\u307e\u3059\u3002Keras\u3092\u4f7f\u3044\u3053\u306a\u305b\u3070\u3001\u6a5f\u68b0\u5b66\u7fd2\u30a8\u30f3\u30b8\u30cb\u30a2\u3068\u3057\u3066\u306e\u30ad\u30e3\u30ea\u30a2\u30a2\u30c3\u30d7\u3082\u5922\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002\u3055\u3063\u305d\u304fKeras\u306e\u4e16\u754c\u306b\u98db\u3073\u8fbc\u3093\u3067\u307f\u307e\u3057\u3087\u3046\uff01<\/p>\n\n\n\n
\u76ee\u6b21 <\/p>\n
Keras\u306f\u3001TensorFlow\u3084Theano\u306a\u3069\u306e\u6df1\u5c64\u5b66\u7fd2\u30e9\u30a4\u30d6\u30e9\u30ea\u306e\u4e0a\u306b\u69cb\u7bc9\u3055\u308c\u305f\u3001\u9ad8\u6c34\u6e96\u306e\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u30e9\u30a4\u30d6\u30e9\u30ea\u3067\u3059\u3002Pyth\u3067\u66f8\u304b\u308c\u3066\u304a\u308a\u3001\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u8fc5\u901f\u306a\u30d7\u30ed\u30c8\u30bf\u30a4\u30d4\u30f3\u30b0\u3092\u53ef\u80fd\u306b\u3057\u307e\u3059\u3002\u30b7\u30f3\u30d7\u30eb\u3067\u76f4\u611f\u7684\u306aAPI\u3092\u63d0\u4f9b\u3057\u3001\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u69cb\u7bc9\u3084\u8a13\u7df4\u3092\u5bb9\u6613\u306b\u3059\u308b\u306e\u304c\u7279\u5fb4\u3067\u3059\u3002<\/p>\n\n\n\n
Keras\u306e\u4e3b\u306a\u7279\u5fb4\u306f\u4ee5\u4e0b\u306e\u901a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n
Keras\u3092\u4f7f\u3048\u3070\u3001\u8fc5\u901f\u306a\u30d7\u30ed\u30c8\u30bf\u30a4\u30d4\u30f3\u30b0\u3068\u5b9f\u9a13\u304c\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\u3002\u30b3\u30fc\u30c9\u306e\u8a18\u8ff0\u91cf\u304c\u5c11\u306a\u304f\u3066\u6e08\u3080\u305f\u3081\u3001\u751f\u7523\u6027\u304c\u9ad8\u3044\u306e\u3082\u9b45\u529b\u3067\u3059\u3002\u521d\u5fc3\u8005\u306b\u3082\u6271\u3044\u3084\u3059\u304f\u3001\u6a5f\u68b0\u5b66\u7fd2\u306e\u6577\u5c45\u3092\u4e0b\u3052\u3066\u304f\u308c\u307e\u3059\u3002\u307e\u305f\u3001\u5927\u898f\u6a21\u306a\u30b3\u30df\u30e5\u30cb\u30c6\u30a3\u3068\u30a8\u30b3\u30b7\u30b9\u30c6\u30e0\u304c\u3042\u308a\u3001\u8c4a\u5bcc\u306a\u30ea\u30bd\u30fc\u30b9\u3084\u60c5\u5831\u304c\u5229\u7528\u53ef\u80fd\u3067\u3059\u3002<\/p>\n\n\n\n
Keras\u3092\u4f7f\u3044\u59cb\u3081\u308b\u306b\u306f\u3001\u307e\u305aPython\u306e\u74b0\u5883\u3092\u6574\u3048\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\u3002Python 3.7\u4ee5\u964d\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u63a8\u5968\u3057\u307e\u3059\u3002\u6b21\u306b\u3001pip\u3092\u4f7f\u3063\u3066Keras\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
pip install keras<\/pre>\n\n\n\nGPU\u3092\u4f7f\u7528\u3059\u308b\u5834\u5408\u306f\u3001\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u3068\u3057\u3066TensorFlow\u3092\u9078\u629e\u3057\u3001CUDA Toolkit\u3068cuDNN\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u8a2d\u5b9a\u304c\u5b8c\u4e86\u3059\u308c\u3070\u3001Keras\u3092\u4f7f\u3063\u305f\u958b\u767a\u306e\u6e96\u5099\u306f\u6574\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
Anaconda\u3092\u4f7f\u3063\u3066\u3044\u308b\u5834\u5408\u306f\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306bconda\u30b3\u30de\u30f3\u30c9\u3067Keras\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
conda install -c conda-forge keras<\/pre>\n\n\n\n\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u304c\u5b8c\u4e86\u3057\u305f\u3089\u3001Python\u30a4\u30f3\u30bf\u30d7\u30ea\u30bf\u3084\u30ce\u30fc\u30c8\u30d6\u30c3\u30af\u74b0\u5883\uff08Jupyter Notebook\u3084Google Colab\uff09\u3092\u8d77\u52d5\u3057\u3066\u3001Keras\u304c\u6b63\u3057\u304f\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3055\u308c\u3066\u3044\u308b\u304b\u78ba\u8a8d\u3057\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
import keras\nkeras.__version__<\/pre>\n\n\n\n\u30d0\u30fc\u30b8\u30e7\u30f3\u60c5\u5831\u304c\u8868\u793a\u3055\u308c\u308c\u3070\u3001Keras\u3092\u4f7f\u3046\u6e96\u5099\u306f\u5b8c\u4e86\u3067\u3059\u3002\u3053\u308c\u3067\u3001\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u3068\u8a13\u7df4\u3092\u59cb\u3081\u3089\u308c\u307e\u3059\u3002\u6b21\u306f\u3001Keras\u306e\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3092\u898b\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
Keras\u3067\u7c21\u5358\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u69cb\u7bc9\uff01\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3092\u5fb9\u5e95\u89e3\u8aac<\/h2>\n\n\n\n
\u30b7\u30fc\u30b1\u30f3\u30b7\u30e3\u30eb\u30e2\u30c7\u30eb\u306e\u4f5c\u308a\u65b9<\/h3>\n\n\n\n
Keras\u3067\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3059\u308b\u969b\u3001\u3082\u3063\u3068\u3082\u57fa\u672c\u7684\u306a\u65b9\u6cd5\u306fSequential\u30e2\u30c7\u30eb\u3092\u4f7f\u3046\u65b9\u6cd5\u3067\u3059\u3002Sequential\u30af\u30e9\u30b9\u3092\u4f7f\u3063\u3066\u3001\u30ec\u30a4\u30e4\u30fc\u3092\u9806\u756a\u306b\u7a4d\u307f\u91cd\u306d\u308b\u3053\u3068\u3067\u30e2\u30c7\u30eb\u3092\u5b9a\u7fa9\u3057\u307e\u3059\u3002add()\u30e1\u30bd\u30c3\u30c9\u3092\u4f7f\u3063\u3066\u3001\u5fc5\u8981\u306a\u30ec\u30a4\u30e4\u30fc\u3092\u8ffd\u52a0\u3057\u3066\u3044\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
\u3088\u304f\u4f7f\u308f\u308c\u308b\u30ec\u30a4\u30e4\u30fc\u306b\u306f\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u3082\u306e\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
\u3053\u308c\u3089\u306e\u30ec\u30a4\u30e4\u30fc\u3092\u7d44\u307f\u5408\u308f\u305b\u3066\u3001\u76ee\u7684\u306b\u5408\u3063\u305f\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08\u30b7\u30fc\u30b1\u30f3\u30b7\u30e3\u30eb\u30e2\u30c7\u30eb\u306e\u4f8b\uff09:<\/p>\n\n\n\n
from keras.models import Sequential\nfrom keras.layers import Dense, Activation, Dropout\n\nmodel = Sequential()\nmodel.add(Dense(64, input_shape=(100,)))\nmodel.add(Activation('relu'))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(10))\nmodel.add(Activation('softmax'))<\/pre>\n\n\n\n\u3053\u306e\u4f8b\u3067\u306f\u3001\u5165\u529b\u30b5\u30a4\u30ba\u304c100\u3001\u96a0\u308c\u5c64\u306e\u30ce\u30fc\u30c9\u6570\u304c64\u3001\u51fa\u529b\u5c64\u306e\u30ce\u30fc\u30c9\u6570\u304c10\u306e\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3057\u3066\u3044\u307e\u3059\u3002\u96a0\u308c\u5c64\u306e\u6d3b\u6027\u5316\u95a2\u6570\u306b\u306fReLU\u95a2\u6570\u3092\u3001\u51fa\u529b\u5c64\u306b\u306f\u30bd\u30d5\u30c8\u30de\u30c3\u30af\u30b9\u95a2\u6570\u3092\u4f7f\u7528\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u6d3b\u6027\u5316\u95a2\u6570\u3084\u640d\u5931\u95a2\u6570\u306e\u9078\u3073\u65b9<\/h3>\n\n\n\n
\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3059\u308b\u969b\u3001\u6d3b\u6027\u5316\u95a2\u6570\u3068\u640d\u5931\u95a2\u6570\u306e\u9078\u629e\u306f\u91cd\u8981\u306a\u30dd\u30a4\u30f3\u30c8\u3067\u3059\u3002<\/p>\n\n\n\n
\u6d3b\u6027\u5316\u95a2\u6570\u306f\u3001\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u5404\u30ce\u30fc\u30c9\u306e\u51fa\u529b\u3092\u6c7a\u5b9a\u3059\u308b\u95a2\u6570\u3067\u3059\u3002\u3088\u304f\u4f7f\u308f\u308c\u308b\u6d3b\u6027\u5316\u95a2\u6570\u306b\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u3082\u306e\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
\u640d\u5931\u95a2\u6570\u306f\u3001\u30e2\u30c7\u30eb\u306e\u4e88\u6e2c\u5024\u3068\u5b9f\u969b\u306e\u5024\u306e\u5dee\u3092\u6e2c\u5b9a\u3057\u3001\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u3092\u8a55\u4fa1\u3057\u307e\u3059\u3002\u4f7f\u7528\u3059\u308b\u640d\u5931\u95a2\u6570\u306f\u3001\u89e3\u6c7a\u3057\u3088\u3046\u3068\u3057\u3066\u3044\u308b\u554f\u984c\u306b\u3088\u3063\u3066\u7570\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
\u30b7\u30fc\u30b1\u30f3\u30b7\u30e3\u30eb\u30e2\u30c7\u30eb\u3092\u5b9a\u7fa9\u3057\u305f\u3089\u3001compile()\u30e1\u30bd\u30c3\u30c9\u3092\u4f7f\u3063\u3066\u30e2\u30c7\u30eb\u3092\u30b3\u30f3\u30d1\u30a4\u30eb\u3057\u307e\u3059\u3002\u3053\u306e\u969b\u3001\u6700\u9069\u5316\u30a2\u30eb\u30b4\u30ea\u30ba\u30e0\uff08optimizer\uff09\u3001\u640d\u5931\u95a2\u6570\uff08loss\uff09\u3001\u8a55\u4fa1\u6307\u6a19\uff08metrics\uff09\u3092\u6307\u5b9a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
model.compile(optimizer='adam',\n loss='categorical_crossentropy',\n metrics=['accuracy'])<\/pre>\n\n\n\n\u3053\u306e\u4f8b\u3067\u306f\u3001\u6700\u9069\u5316\u30a2\u30eb\u30b4\u30ea\u30ba\u30e0\u306bAdam\u3001\u640d\u5931\u95a2\u6570\u306b\u30ab\u30c6\u30b4\u30ea\u30ab\u30eb\u30af\u30ed\u30b9\u30a8\u30f3\u30c8\u30ed\u30d4\u30fc\u3001\u8a55\u4fa1\u6307\u6a19\u306b\u7cbe\u5ea6\uff08accuracy\uff09\u3092\u4f7f\u7528\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u30b3\u30f3\u30d1\u30a4\u30eb\u5f8c\u3001fit()\u30e1\u30bd\u30c3\u30c9\u3092\u4f7f\u3063\u3066\u30e2\u30c7\u30eb\u3092\u8a13\u7df4\u30c7\u30fc\u30bf\u3067\u5b66\u7fd2\u3055\u305b\u307e\u3059\u3002\u30a8\u30dd\u30c3\u30af\u6570\uff08epochs\uff09\u3068\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba\uff08batch_size\uff09\u3092\u8a2d\u5b9a\u3057\u3066\u3001\u5b66\u7fd2\u3092\u5b9f\u884c\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
model.fit(x_train, y_train, epochs=10, batch_size=32)<\/pre>\n\n\n\n\u3053\u306e\u4f8b\u3067\u306f\u3001\u8a13\u7df4\u30c7\u30fc\u30bf\uff08x_train, y_train\uff09\u3092\u4f7f\u3063\u3066\u300110\u30a8\u30dd\u30c3\u30af\u3001\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba32\u3067\u30e2\u30c7\u30eb\u3092\u5b66\u7fd2\u3055\u305b\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u5b66\u7fd2\u304c\u5b8c\u4e86\u3057\u305f\u3089\u3001evaluate()\u30e1\u30bd\u30c3\u30c9\u3067\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u3092\u4f7f\u3063\u3066\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u3092\u8a55\u4fa1\u3057\u305f\u308a\u3001predict()\u30e1\u30bd\u30c3\u30c9\u3067\u65b0\u3057\u3044\u30c7\u30fc\u30bf\u306b\u5bfe\u3059\u308b\u4e88\u6e2c\u3092\u884c\u3063\u305f\u308a\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
\u4ee5\u4e0a\u304c\u3001Keras\u3067\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3057\u3001\u5b66\u7fd2\u3055\u305b\u308b\u307e\u3067\u306e\u4e00\u9023\u306e\u6d41\u308c\u3067\u3059\u3002\u30b7\u30f3\u30d7\u30eb\u3067\u76f4\u611f\u7684\u306a\u30b3\u30fc\u30c9\u3067\u3001\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u304b\u3089\u8a13\u7df4\u307e\u3067\u3092\u884c\u3048\u308b\u306e\u304cKeras\u306e\u5927\u304d\u306a\u9b45\u529b\u3067\u3059\u3002<\/p>\n\n\n\n
Keras \u304c\u3055\u3055\u3048\u308b\u6700\u5148\u7aef\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3 \u2013 CNN \u3068 RNN \u306e\u5b9f\u88c5\u4f8b<\/h2>\n\n\n\n
Keras\u306f\u3001CNN\u3084RNN\u306a\u3069\u306e\u6700\u5148\u7aef\u306e\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3092\u7c21\u5358\u306b\u5b9f\u88c5\u3067\u304d\u308b\u6a5f\u80fd\u3092\u63d0\u4f9b\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u3053\u3067\u306f\u3001Keras\u3067CNN\u3068RNN\u3092\u69cb\u7bc9\u3059\u308b\u65b9\u6cd5\u3092\u898b\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
Keras \u3067\u306e CNN \u306e\u69cb\u7bc9\u65b9\u6cd5<\/h3>\n\n\n\n
CNN\u306f\u3001\u753b\u50cf\u30c7\u30fc\u30bf\u306e\u7279\u5fb4\u62bd\u51fa\u3084\u5206\u985e\u306b\u512a\u308c\u305f\u6027\u80fd\u3092\u767a\u63ee\u3057\u307e\u3059\u3002Keras\u3067CNN\u3092\u69cb\u7bc9\u3059\u308b\u306b\u306f\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u30ec\u30a4\u30e4\u30fc\u3092\u4f7f\u7528\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u3053\u308c\u3089\u306e\u30ec\u30a4\u30e4\u30fc\u3092Sequential\u30e2\u30c7\u30eb\u307e\u305f\u306fFunctional API\u3092\u4f7f\u3063\u3066\u7d44\u307f\u5408\u308f\u305b\u3001CNN\u3092\u69cb\u7bc9\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08CNN\u306e\u4f8b\uff09:<\/p>\n\n\n\n
from keras.models import Sequential\nfrom keras.layers import Conv2D, MaxPooling2D, Flatten, Dense\n\nmodel = Sequential()\nmodel.add(Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))\nmodel.add(MaxPooling2D((2, 2)))\nmodel.add(Conv2D(64, (3, 3), activation='relu'))\nmodel.add(MaxPooling2D((2, 2)))\nmodel.add(Conv2D(64, (3, 3), activation='relu'))\nmodel.add(Flatten())\nmodel.add(Dense(64, activation='relu'))\nmodel.add(Dense(10, activation='softmax'))<\/pre>\n\n\n\n\u3053\u306e\u4f8b\u3067\u306f\u3001MNIST\u306e\u624b\u66f8\u304d\u6570\u5b57\u5206\u985e\u30bf\u30b9\u30af\u3092\u60f3\u5b9a\u3057\u305f\u7c21\u5358\u306aCNN\u3092\u69cb\u7bc9\u3057\u3066\u3044\u307e\u3059\u3002\u7573\u307f\u8fbc\u307f\u5c64\u3068\u30de\u30c3\u30af\u30b9\u30d7\u30fc\u30ea\u30f3\u30b0\u5c64\u3092\u4ea4\u4e92\u306b\u91cd\u306d\u308b\u3053\u3068\u3067\u3001\u753b\u50cf\u306e\u7279\u5fb4\u3092\u968e\u5c64\u7684\u306b\u62bd\u51fa\u3057\u3001\u6700\u5f8c\u306b\u5168\u7d50\u5408\u5c64\u3067\u5206\u985e\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
Keras \u3067\u306e RNN\u30fbLSTM \u306e\u5229\u7528\u65b9\u6cd5<\/h3>\n\n\n\n
RNN\u306f\u3001\u6642\u7cfb\u5217\u30c7\u30fc\u30bf\u3084\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\u30bf\u30b9\u30af\u3067\u5a01\u529b\u3092\u767a\u63ee\u3057\u307e\u3059\u3002Keras\u3067RNN\u3092\u69cb\u7bc9\u3059\u308b\u969b\u306f\u3001\u4ee5\u4e0b\u306e\u30ec\u30a4\u30e4\u30fc\u3092\u4f7f\u7528\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u3053\u308c\u3089\u306e\u30ec\u30a4\u30e4\u30fc\u3092\u7d44\u307f\u5408\u308f\u305b\u3066RNN\u3092\u69cb\u7bc9\u3057\u307e\u3059\u3002\u5165\u529b\u30c7\u30fc\u30bf\u3092\u9069\u5207\u306a\u5f62\u72b6\u306b\u5909\u63db\u3057\u3001Embedding\u30ec\u30a4\u30e4\u30fc\u3092\u4f7f\u3063\u3066\u5358\u8a9e\u3092\u30d9\u30af\u30c8\u30eb\u5316\u3059\u308b\u306e\u304c\u30dd\u30a4\u30f3\u30c8\u3067\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08LSTM\u306e\u4f8b\uff09:<\/p>\n\n\n\n
from keras.models import Sequential\nfrom keras.layers import LSTM, Dense, Embedding\n\nmodel = Sequential()\nmodel.add(Embedding(input_dim=max_words, output_dim=32, input_length=max_len))\nmodel.add(LSTM(32))\nmodel.add(Dense(1, activation='sigmoid'))<\/pre>\n\n\n\n\u3053\u306e\u4f8b\u3067\u306f\u3001\u6620\u753b\u30ec\u30d3\u30e5\u30fc\u306e\u611f\u60c5\u5206\u6790\u30bf\u30b9\u30af\u3092\u60f3\u5b9a\u3057\u305f\u7c21\u5358\u306aLSTM\u30e2\u30c7\u30eb\u3092\u69cb\u7bc9\u3057\u3066\u3044\u307e\u3059\u3002Embedding\u30ec\u30a4\u30e4\u30fc\u3067\u5358\u8a9e\u3092\u30d9\u30af\u30c8\u30eb\u5316\u3057\u3001LSTM\u30ec\u30a4\u30e4\u30fc\u3067\u6642\u7cfb\u5217\u30c7\u30fc\u30bf\u3092\u51e6\u7406\u3057\u305f\u5f8c\u3001\u5168\u7d50\u5408\u5c64\u3067\u4e8c\u5024\u5206\u985e\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u8abf\u6574\u3068\u30e2\u30c7\u30eb\u306e\u8a55\u4fa1<\/h3>\n\n\n\n
CNN\u3084RNN\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u6700\u5927\u9650\u306b\u5f15\u304d\u51fa\u3059\u306b\u306f\u3001\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u9069\u5207\u306a\u8abf\u6574\u304c\u6b20\u304b\u305b\u307e\u305b\u3093\u3002\u91cd\u8981\u306a\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u306b\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u3082\u306e\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
\u3053\u308c\u3089\u306e\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u3092\u624b\u52d5\u3067\u8abf\u6574\u3059\u308b\u306e\u306f\u5927\u5909\u306a\u306e\u3067\u3001\u30b0\u30ea\u30c3\u30c9\u30b5\u30fc\u30c1\u3084\u30e9\u30f3\u30c0\u30e0\u30b5\u30fc\u30c1\u3092\u4f7f\u3063\u3066\u81ea\u52d5\u7684\u306b\u6700\u9069\u306a\u7d44\u307f\u5408\u308f\u305b\u3092\u63a2\u7d22\u3059\u308b\u306e\u304c\u4e00\u822c\u7684\u3067\u3059\u3002<\/p>\n\n\n\n
\u307e\u305f\u3001\u30e2\u30c7\u30eb\u306e\u6c4e\u5316\u6027\u80fd\u3092\u8a55\u4fa1\u3059\u308b\u305f\u3081\u306b\u3001\u4ea4\u5dee\u691c\u8a3c\u3092\u884c\u3046\u306e\u304c\u671b\u307e\u3057\u3044\u3067\u3059\u3002\u30c7\u30fc\u30bf\u3092\u8a13\u7df4\u30bb\u30c3\u30c8\u3068\u30c6\u30b9\u30c8\u30bb\u30c3\u30c8\u306b\u5206\u5272\u3057\u3001\u8a13\u7df4\u30bb\u30c3\u30c8\u3067\u30e2\u30c7\u30eb\u3092\u5b66\u7fd2\u3055\u305b\u305f\u5f8c\u3001\u30c6\u30b9\u30c8\u30bb\u30c3\u30c8\u3067\u6027\u80fd\u3092\u8a55\u4fa1\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u8a55\u4fa1\u6307\u6a19\u3068\u3057\u3066\u306f\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u3082\u306e\u304c\u3088\u304f\u4f7f\u308f\u308c\u307e\u3059\u3002<\/p>\n\n\n\n
\u3053\u308c\u3089\u306e\u6307\u6a19\u3092\u8a08\u7b97\u3059\u308b\u3053\u3068\u3067\u3001\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u3092\u591a\u89d2\u7684\u306b\u8a55\u4fa1\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
Keras\u3092\u4f7f\u3048\u3070\u3001\u6700\u5148\u7aef\u306eCNN\u3084RNN\u3092\u7c21\u5358\u306b\u69cb\u7bc9\u3057\u3001\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u8abf\u6574\u3084\u8a55\u4fa1\u3092\u884c\u3048\u307e\u3059\u3002\u305c\u3072\u5b9f\u969b\u306b\u30b3\u30fc\u30c9\u3092\u66f8\u3044\u3066\u3001\u305d\u306e\u529b\u3092\u4f53\u611f\u3057\u3066\u307f\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n
Keras\u306f\u6a5f\u68b0\u5b66\u7fd2\u306e\u5206\u91ce\u3067\u5e83\u304f\u4f7f\u308f\u308c\u3066\u3044\u308b\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u304c\u3001TensorFlow\u3084PyTorch\u3068\u3044\u3063\u305f\u4ed6\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u306e\u9055\u3044\u3092\u7406\u89e3\u3059\u308b\u3053\u3068\u304c\u91cd\u8981\u3067\u3059\u3002\u3053\u3053\u3067\u306f\u3001Keras\u3068TensorFlow\u3001PyTorch\u306e\u6bd4\u8f03\u3092\u901a\u3058\u3066\u3001Keras\u306e\u7279\u5fb4\u3084\u4f7f\u3044\u6240\u3092\u63a2\u3063\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
TensorFlow\u306fGoogle\u304c\u958b\u767a\u3057\u305f\u6a5f\u68b0\u5b66\u7fd2\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002Keras\u306f\u5f53\u521d\u3001TensorFlow\u4e0a\u3067\u52d5\u4f5c\u3059\u308b\u9ad8\u30ec\u30d9\u30ebAPI\u3068\u3057\u3066\u958b\u767a\u3055\u308c\u307e\u3057\u305f\u3002TensorFlow 2.0\u304b\u3089\u306f\u3001Keras\u304cTensorFlow\u306e\u30c7\u30d5\u30a9\u30eb\u30c8\u306e\u9ad8\u30ec\u30d9\u30ebAPI\u3068\u3057\u3066\u7d71\u5408\u3055\u308c\u3066\u3044\u307e\u3059\u3002TensorFlow\u306f\u4f4e\u30ec\u30d9\u30eb\u306e\u5236\u5fa1\u304c\u53ef\u80fd\u3067\u67d4\u8edf\u6027\u304c\u9ad8\u3044\u4e00\u65b9\u3001Keras\u306f\u30b7\u30f3\u30d7\u30eb\u3067\u4f7f\u3044\u3084\u3059\u3044\u306e\u304c\u7279\u5fb4\u3067\u3059\u3002<\/p>\n\n\n\n
PyTorch\u306fFacebook\u304c\u958b\u767a\u3057\u305f\u6a5f\u68b0\u5b66\u7fd2\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002PyTorch\u306f\u52d5\u7684\u8a08\u7b97\u30b0\u30e9\u30d5\u3092\u63a1\u7528\u3057\u3066\u3044\u308b\u305f\u3081\u3001\u30c7\u30d0\u30c3\u30b0\u3084\u30e2\u30c7\u30eb\u306e\u4fee\u6b63\u304c\u5bb9\u6613\u3067\u3059\u3002\u4e00\u65b9\u3001Keras\u306f\u9759\u7684\u8a08\u7b97\u30b0\u30e9\u30d5\u3092\u4f7f\u7528\u3057\u3066\u304a\u308a\u3001\u30e2\u30c7\u30eb\u306e\u5b9a\u7fa9\u304c\u30b7\u30f3\u30d7\u30eb\u3067\u5206\u304b\u308a\u3084\u3059\u304f\u306a\u3063\u3066\u3044\u307e\u3059\u3002PyTorch\u306f\u3088\u308a\u67d4\u8edf\u6027\u304c\u9ad8\u304f\u3001\u7814\u7a76\u7528\u9014\u306b\u9069\u3057\u3066\u3044\u307e\u3059\u304c\u3001Keras\u306f\u7c21\u6f54\u3055\u3068\u4f7f\u3044\u3084\u3059\u3055\u306b\u512a\u308c\u3001\u30d7\u30ed\u30c0\u30af\u30b7\u30e7\u30f3\u74b0\u5883\u306b\u9069\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08Keras\u3068PyTorch\u306e\u6bd4\u8f03\uff09:<\/p>\n\n\n\n
Keras:<\/p>\n\n\n\n
from keras.models import Sequential\nfrom keras.layers import Dense\n\nmodel = Sequential()\nmodel.add(Dense(64, activation='relu', input_shape=(100,)))\nmodel.add(Dense(10, activation='softmax'))\n\nmodel.compile(optimizer='adam',\n loss='categorical_crossentropy',\n metrics=['accuracy'])<\/pre>\n\n\n\nPyTorch:<\/p>\n\n\n\n
import torch.nn as nn\n\nclass Model(nn.Module):\n def __init__(self):\n super(Model, self).__init__()\n self.fc1 = nn.Linear(100, 64)\n self.fc2 = nn.Linear(64, 10)\n\n def forward(self, x):\n x = nn.functional.relu(self.fc1(x))\n x = self.fc2(x)\n return nn.functional.softmax(x, dim=1)\n\nmodel = Model()\n\ncriterion = nn.CrossEntropyLoss()\noptimizer = torch.optim.Adam(model.parameters())<\/pre>\n\n\n\n\u4e0a\u8a18\u306e\u30b3\u30fc\u30c9\u4f8b\u304b\u3089\u3082\u5206\u304b\u308b\u3088\u3046\u306b\u3001Keras\u3067\u306f\u30b7\u30f3\u30d7\u30eb\u3067\u8aad\u307f\u3084\u3059\u3044\u30b3\u30fc\u30c9\u3067\u30e2\u30c7\u30eb\u3092\u5b9a\u7fa9\u3067\u304d\u307e\u3059\u3002\u4e00\u65b9\u3001PyTorch\u3067\u306f\u3088\u308a\u67d4\u8edf\u306a\u30e2\u30c7\u30eb\u5b9a\u7fa9\u304c\u53ef\u80fd\u3067\u3059\u304c\u3001\u30b3\u30fc\u30c9\u304c\u82e5\u5e72\u8907\u96d1\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
Keras \u3092\u4f7f\u3046\u30e1\u30ea\u30c3\u30c8\u3068\u30c7\u30e1\u30ea\u30c3\u30c8<\/h3>\n\n\n\n
Keras\u3092\u4f7f\u3046\u30e1\u30ea\u30c3\u30c8\u306f\u3001\u4f55\u3068\u3044\u3063\u3066\u3082\u7c21\u6f54\u3067\u8aad\u307f\u3084\u3059\u3044\u30b3\u30fc\u30c9\u3067\u30e2\u30c7\u30eb\u3092\u5b9a\u7fa9\u3067\u304d\u308b\u3053\u3068\u3067\u3059\u3002\u4e8b\u524d\u306b\u5b9a\u7fa9\u3055\u308c\u305f\u30ec\u30a4\u30e4\u30fc\u3084\u640d\u5931\u95a2\u6570\u3001\u6700\u9069\u5316\u30a2\u30eb\u30b4\u30ea\u30ba\u30e0\u3092\u7d44\u307f\u5408\u308f\u305b\u308b\u3060\u3051\u3067\u3001\u7c21\u5358\u306b\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3067\u304d\u307e\u3059\u3002\u307e\u305f\u3001\u30de\u30eb\u30c1GPU\u3084TPU\u3078\u306e\u5bfe\u5fdc\u306a\u3069\u3001\u30b9\u30b1\u30fc\u30e9\u30d3\u30ea\u30c6\u30a3\u304c\u9ad8\u3044\u306e\u3082\u5927\u304d\u306a\u5229\u70b9\u3067\u3059\u3002\u8c4a\u5bcc\u306a\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3084\u30b3\u30df\u30e5\u30cb\u30c6\u30a3\u306b\u3088\u308b\u30b5\u30dd\u30fc\u30c8\u3082\u3001Keras\u3092\u4f7f\u3046\u4e0a\u3067\u306e\u5fc3\u5f37\u3044\u5473\u65b9\u3067\u3059\u3002<\/p>\n\n\n\n
\u4e00\u65b9\u3001Keras\u3092\u4f7f\u3046\u30c7\u30e1\u30ea\u30c3\u30c8\u3068\u3057\u3066\u306f\u3001\u4f4e\u30ec\u30d9\u30eb\u306e\u5236\u5fa1\u304c\u96e3\u3057\u304f\u3001\u30ab\u30b9\u30bf\u30de\u30a4\u30ba\u306e\u81ea\u7531\u5ea6\u304c\u4f4e\u3044\u3053\u3068\u304c\u6319\u3052\u3089\u308c\u307e\u3059\u3002\u30c7\u30d0\u30c3\u30b0\u3084\u30a8\u30e9\u30fc\u51e6\u7406\u304c\u96e3\u3057\u3044\u5834\u5408\u3082\u3042\u308a\u307e\u3059\u3002\u8907\u96d1\u306a\u30e2\u30c7\u30eb\u306e\u5b9f\u88c5\u306b\u306f\u5411\u3044\u3066\u3044\u306a\u3044\u53ef\u80fd\u6027\u304c\u3042\u308a\u3001\u7814\u7a76\u7528\u9014\u3067\u306f\u67d4\u8edf\u6027\u304c\u8db3\u308a\u306a\u3044\u3053\u3068\u3082\u3042\u308b\u3067\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
\u305f\u3060\u3057\u3001\u3053\u308c\u3089\u306e\u30c7\u30e1\u30ea\u30c3\u30c8\u306fKeras\u306e\u8a2d\u8a08\u601d\u60f3\u306b\u3088\u308b\u3082\u306e\u3067\u3042\u308a\u3001\u30b7\u30f3\u30d7\u30eb\u3055\u3068\u4f7f\u3044\u3084\u3059\u3055\u3092\u91cd\u8996\u3059\u308b\u5834\u5408\u306b\u306f\u3001\u3080\u3057\u308d\u30e1\u30ea\u30c3\u30c8\u3068\u8a00\u3048\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
Keras\u306f\u7c21\u5358\u3055\u3068\u67d4\u8edf\u6027\u306e\u30d0\u30e9\u30f3\u30b9\u304c\u7d76\u5999\u306a\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002\u521d\u5fc3\u8005\u304b\u3089\u7d4c\u9a13\u8005\u307e\u3067\u3001\u5e45\u5e83\u3044\u30e6\u30fc\u30b6\u30fc\u306e\u30cb\u30fc\u30ba\u306b\u5fdc\u3048\u3089\u308c\u308b\u3001\u6c4e\u7528\u6027\u306e\u9ad8\u3044\u30c4\u30fc\u30eb\u3060\u3068\u8a00\u3048\u308b\u3067\u3057\u3087\u3046\u3002TensorFlow\u3084PyTorch\u3068\u306e\u4f7f\u3044\u5206\u3051\u3092\u7406\u89e3\u3057\u3001\u9069\u6750\u9069\u6240\u3067\u6d3b\u7528\u3059\u308b\u3053\u3068\u304c\u91cd\u8981\u3067\u3059\u3002<\/p>\n\n\n\n
Keras \u3067\u6a5f\u68b0\u5b66\u7fd2\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u9032\u3081\u308b\u30b3\u30c4<\/h2>\n\n\n\n
Keras\u3092\u4f7f\u3063\u3066\u6a5f\u68b0\u5b66\u7fd2\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u9032\u3081\u308b\u969b\u306b\u306f\u3001\u3044\u304f\u3064\u304b\u306e\u91cd\u8981\u306a\u30dd\u30a4\u30f3\u30c8\u3092\u62bc\u3055\u3048\u3066\u304a\u304f\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\u3002\u3053\u3053\u3067\u306f\u3001\u30c7\u30fc\u30bf\u306e\u6e96\u5099\u304b\u3089\u7d50\u679c\u306e\u5206\u6790\u307e\u3067\u3001\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u6210\u529f\u306b\u5c0e\u304f\u305f\u3081\u306e\u30b3\u30c4\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u30c7\u30fc\u30bf\u306e\u6e96\u5099\u3068\u524d\u51e6\u7406\u306e\u30dd\u30a4\u30f3\u30c8<\/h3>\n\n\n\n
\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u306f\u3001\u5165\u529b\u3055\u308c\u308b\u30c7\u30fc\u30bf\u306e\u8cea\u306b\u5927\u304d\u304f\u4f9d\u5b58\u3057\u307e\u3059\u3002\u305d\u306e\u305f\u3081\u3001\u9ad8\u54c1\u8cea\u306a\u30c7\u30fc\u30bf\u3092\u6e96\u5099\u3059\u308b\u3053\u3068\u304c\u4f55\u3088\u308a\u3082\u91cd\u8981\u3067\u3059\u3002\u6b20\u640d\u5024\u3084\u7570\u5e38\u5024\u306e\u51e6\u7406\u3001\u30c7\u30fc\u30bf\u306e\u6b63\u898f\u5316\u306a\u3069\u306e\u524d\u51e6\u7406\u3092\u9069\u5207\u306b\u884c\u3044\u307e\u3057\u3087\u3046\u3002\u7279\u5fb4\u91cf\u30a8\u30f3\u30b8\u30cb\u30a2\u30ea\u30f3\u30b0\u3092\u65bd\u3059\u3053\u3068\u3067\u3001\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u3092\u5927\u5e45\u306b\u5411\u4e0a\u3055\u305b\u3089\u308c\u308b\u5834\u5408\u3082\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
\u30c7\u30fc\u30bf\u306e\u53d6\u308a\u6271\u3044\u306b\u306f\u3001NumPy\u3084Pandas\u306a\u3069\u306e\u30e9\u30a4\u30d6\u30e9\u30ea\u3092\u6d3b\u7528\u3059\u308b\u3068\u4fbf\u5229\u3067\u3059\u3002\u4ee5\u4e0b\u306f\u3001scikit-learn\u3092\u4f7f\u3063\u3066\u30c7\u30fc\u30bf\u3092\u6b63\u898f\u5316\u3059\u308b\u4f8b\u3067\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08\u30c7\u30fc\u30bf\u306e\u6b63\u898f\u5316\uff09:<\/p>\n\n\n\n
from sklearn.preprocessing import StandardScaler\n\nscaler = StandardScaler()\nX_train = scaler.fit_transform(X_train)\nX_test = scaler.transform(X_test)<\/pre>\n\n\n\n\u3053\u306e\u3088\u3046\u306b\u3001\u8a13\u7df4\u30c7\u30fc\u30bf\u3092\u7528\u3044\u3066\u6b63\u898f\u5316\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u3092\u5b66\u7fd2\u3057\u3001\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u306b\u5bfe\u3057\u3066\u306f\u5b66\u7fd2\u6e08\u307f\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u3092\u9069\u7528\u3059\u308b\u3053\u3068\u304c\u91cd\u8981\u3067\u3059\u3002<\/p>\n\n\n\n
\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u63a2\u7d22\u3068\u7d50\u679c\u306e\u5206\u6790\u65b9\u6cd5<\/h3>\n\n\n\n
\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306f\u3001\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u8a2d\u5b9a\u306b\u5927\u304d\u304f\u5de6\u53f3\u3055\u308c\u307e\u3059\u3002\u30b0\u30ea\u30c3\u30c9\u30b5\u30fc\u30c1\u3084\u30e9\u30f3\u30c0\u30e0\u30b5\u30fc\u30c1\u3092\u4f7f\u3063\u3066\u3001\u6700\u9069\u306a\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u7d44\u307f\u5408\u308f\u305b\u3092\u63a2\u7d22\u3057\u307e\u3057\u3087\u3046\u3002\u307e\u305f\u3001\u4ea4\u5dee\u691c\u8a3c\u3092\u884c\u3046\u3053\u3068\u3067\u3001\u30e2\u30c7\u30eb\u306e\u6c4e\u5316\u6027\u80fd\u3092\u6b63\u3057\u304f\u8a55\u4fa1\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
\u4ee5\u4e0b\u306f\u3001Keras\u30e2\u30c7\u30eb\u306b\u5bfe\u3057\u3066\u30b0\u30ea\u30c3\u30c9\u30b5\u30fc\u30c1\u3092\u5b9f\u884c\u3059\u308b\u4f8b\u3067\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08\u30b0\u30ea\u30c3\u30c9\u30b5\u30fc\u30c1\uff09:<\/p>\n\n\n\n
from sklearn.model_selection import GridSearchCV\nfrom keras.wrappers.scikit_learn import KerasClassifier\n\ndef create_model(optimizer='adam'):\n model = Sequential()\n model.add(Dense(64, activation='relu', input_shape=(100,)))\n model.add(Dense(1, activation='sigmoid'))\n model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])\n return model\n\nparam_grid = {'optimizer': ['adam', 'rmsprop', 'sgd']}\nmodel = KerasClassifier(build_fn=create_model, epochs=10, batch_size=32, verbose=0)\ngrid = GridSearchCV(estimator=model, param_grid=param_grid, cv=3)\ngrid_result = grid.fit(X_train, y_train)<\/pre>\n\n\n\n\u3053\u306e\u4f8b\u3067\u306f\u3001\u30aa\u30d7\u30c6\u30a3\u30de\u30a4\u30b6\u306e\u7a2e\u985e\u3092\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u3068\u3057\u3066\u63a2\u7d22\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u30e2\u30c7\u30eb\u306e\u8a13\u7df4\u304c\u7d42\u308f\u3063\u305f\u3089\u3001\u5b66\u7fd2\u66f2\u7dda\u3084\u691c\u8a3c\u66f2\u7dda\u3092\u63cf\u753b\u3057\u3001\u904e\u5b66\u7fd2\u3084\u672a\u5b66\u7fd2\u306e\u5146\u5019\u304c\u306a\u3044\u304b\u30c1\u30a7\u30c3\u30af\u3057\u307e\u3057\u3087\u3046\u3002\u5206\u985e\u554f\u984c\u3067\u3042\u308c\u3070\u3001\u6df7\u540c\u884c\u5217\u3084ROC\u66f2\u7dda\u3092\u4f7f\u3063\u3066\u30e2\u30c7\u30eb\u306e\u6027\u80fd\u3092\u591a\u89d2\u7684\u306b\u5206\u6790\u3059\u308b\u3053\u3068\u3082\u5927\u5207\u3067\u3059\u3002<\/p>\n\n\n\n
\u9665\u308a\u3084\u3059\u3044\u843d\u3068\u3057\u7a74\u3068\u56de\u907f\u65b9\u6cd5<\/h3>\n\n\n\n
\u6a5f\u68b0\u5b66\u7fd2\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u9032\u3081\u308b\u4e0a\u3067\u306f\u3001\u3044\u304f\u3064\u304b\u306e\u843d\u3068\u3057\u7a74\u306b\u6ce8\u610f\u304c\u5fc5\u8981\u3067\u3059\u3002<\/p>\n\n\n\n
\u307e\u305a\u3001\u30c7\u30fc\u30bf\u30ea\u30fc\u30af\u306b\u6c17\u3092\u3064\u3051\u307e\u3057\u3087\u3046\u3002\u8a13\u7df4\u30c7\u30fc\u30bf\u3068\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u3092\u53b3\u5bc6\u306b\u5206\u96e2\u3057\u3001\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u306e\u60c5\u5831\u304c\u8a13\u7df4\u30c7\u30fc\u30bf\u306b\u6df7\u5165\u3057\u306a\u3044\u3088\u3046\u306b\u3059\u308b\u3053\u3068\u304c\u5927\u5207\u3067\u3059\u3002<\/p>\n\n\n\n
\u6b21\u306b\u3001\u904e\u5b66\u7fd2\u3078\u306e\u5bfe\u7b56\u3092\u6020\u3089\u306a\u3044\u3088\u3046\u306b\u3057\u307e\u3057\u3087\u3046\u3002\u6b63\u5247\u5316\u3084\u30c9\u30ed\u30c3\u30d7\u30a2\u30a6\u30c8\u3092\u9069\u5207\u306b\u4f7f\u7528\u3057\u3001\u30e2\u30c7\u30eb\u304c\u8a13\u7df4\u30c7\u30fc\u30bf\u306b\u904e\u5270\u306b\u9069\u5408\u3059\u308b\u306e\u3092\u9632\u304e\u307e\u3059\u3002<\/p>\n\n\n\n
\u307e\u305f\u3001\u30e2\u30c7\u30eb\u306e\u89e3\u91c8\u6027\u3092\u91cd\u8996\u3059\u308b\u3053\u3068\u3082\u5927\u5207\u3067\u3059\u3002\u30d6\u30e9\u30c3\u30af\u30dc\u30c3\u30af\u30b9\u5316\u3057\u305f\u30e2\u30c7\u30eb\u306f\u3001\u5b9f\u969b\u306e\u30d3\u30b8\u30cd\u30b9\u3067\u4f7f\u3046\u306b\u306f\u4e0d\u9069\u5207\u306a\u5834\u5408\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
\u6700\u5f8c\u306b\u3001\u518d\u73fe\u6027\u306e\u78ba\u4fdd\u3092\u5fc3\u304c\u3051\u307e\u3057\u3087\u3046\u3002\u4e71\u6570\u30b7\u30fc\u30c9\u3092\u56fa\u5b9a\u3057\u3001\u5b9f\u9a13\u6761\u4ef6\u3092\u8a18\u9332\u3059\u308b\u3053\u3068\u3067\u3001\u7d50\u679c\u306e\u518d\u73fe\u6027\u3092\u62c5\u4fdd\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
Keras\u3092\u4f7f\u3063\u305f\u6a5f\u68b0\u5b66\u7fd2\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3067\u306f\u3001\u3053\u308c\u3089\u306e\u30dd\u30a4\u30f3\u30c8\u3092\u62bc\u3055\u3048\u308b\u3053\u3068\u304c\u6210\u529f\u3078\u306e\u9375\u3068\u306a\u308a\u307e\u3059\u3002\u30c7\u30fc\u30bf\u306e\u6e96\u5099\u304b\u3089\u7d50\u679c\u306e\u5206\u6790\u307e\u3067\u3001\u4e01\u5be7\u304b\u3064\u6226\u7565\u7684\u306b\u9032\u3081\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
\u3055\u3044\u3054\u306b\uff1aKeras \u3067\u5dee\u3092\u3064\u3051\u308b\u305f\u3081\u306b\u4f55\u3092\u3059\u3079\u304d\u304b\uff1f<\/h2>\n\n\n\n
Keras\u306f\u3001\u6a5f\u68b0\u5b66\u7fd2\u30a8\u30f3\u30b8\u30cb\u30a2\u306b\u3068\u3063\u3066\u7fd2\u5f97\u3059\u3079\u304d\u91cd\u8981\u306a\u30b9\u30ad\u30eb\u306e\u3072\u3068\u3064\u3067\u3059\u3002\u3053\u3053\u3067\u306f\u3001Keras\u3092\u5b66\u3076\u3053\u3068\u306e\u610f\u7fa9\u3068\u3001Keras\u3092\u30de\u30b9\u30bf\u30fc\u3059\u308b\u305f\u3081\u306e\u5b66\u7fd2\u30ed\u30fc\u30c9\u30de\u30c3\u30d7\u3092\u63d0\u6848\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
Keras \u306e\u7fd2\u5f97\u304c\u6a5f\u68b0\u5b66\u7fd2\u30a8\u30f3\u30b8\u30cb\u30a2\u306b\u4e0d\u53ef\u6b20\u306a\u7406\u7531<\/h3>\n\n\n\n
Keras\u306f\u3001\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u3068\u8a13\u7df4\u3092\u7c21\u5358\u306b\u884c\u3048\u308b\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002Keras\u3092\u4f7f\u3044\u3053\u306a\u305b\u308c\u3070\u3001\u751f\u7523\u6027\u3092\u5927\u5e45\u306b\u5411\u4e0a\u3055\u305b\u3089\u308c\u307e\u3059\u3002TensorFlow\u3084PyTorch\u306a\u3069\u306e\u4f4e\u30ec\u30d9\u30eb\u306a\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u6271\u3046\u4e0a\u3067\u3082\u3001Keras\u306e\u77e5\u8b58\u306f\u5f79\u7acb\u3061\u307e\u3059\u3002<\/p>\n\n\n\n
\u307e\u305f\u3001Keras\u3092\u4f7f\u3044\u3053\u306a\u305b\u308b\u3053\u3068\u306f\u3001\u6a5f\u68b0\u5b66\u7fd2\u30a8\u30f3\u30b8\u30cb\u30a2\u3068\u3057\u3066\u306e\u5b9f\u52d9\u80fd\u529b\u3092\u793a\u3059\u3072\u3068\u3064\u306e\u6307\u6a19\u3068\u306a\u308a\u307e\u3059\u3002\u591a\u304f\u306e\u4f01\u696d\u3084\u7d44\u7e54\u3067Keras\u304c\u6d3b\u7528\u3055\u308c\u3066\u304a\u308a\u3001Keras\u306e\u30b9\u30ad\u30eb\u306f\u6c42\u4eba\u8981\u4ef6\u306b\u3082\u306a\u3063\u3066\u3044\u308b\u306e\u3067\u3059\u3002<\/p>\n\n\n\n
Keras\u3092\u7fd2\u5f97\u3059\u308b\u3053\u3068\u306f\u3001\u6a5f\u68b0\u5b66\u7fd2\u30a8\u30f3\u30b8\u30cb\u30a2\u3068\u3057\u3066\u306e\u30ad\u30e3\u30ea\u30a2\u30a2\u30c3\u30d7\u306b\u3064\u306a\u304c\u308a\u307e\u3059\u3002\u6280\u8853\u30c8\u30ec\u30f3\u30c9\u306e\u5909\u5316\u306b\u5bfe\u5fdc\u3057\u3001\u5e38\u306b\u5b66\u3073\u7d9a\u3051\u308b\u59ff\u52e2\u304c\u91cd\u8981\u3067\u3059\u3002<\/p>\n\n\n\n
Keras \u3092\u30de\u30b9\u30bf\u30fc\u3059\u308b\u305f\u3081\u306e\u5b66\u7fd2\u30ed\u30fc\u30c9\u30de\u30c3\u30d7<\/h3>\n\n\n\n
Keras\u3092\u30de\u30b9\u30bf\u30fc\u3059\u308b\u306b\u306f\u3001\u4f53\u7cfb\u7684\u306a\u5b66\u7fd2\u304c\u6b20\u304b\u305b\u307e\u305b\u3093\u3002\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u30b9\u30c6\u30c3\u30d7\u3092\u8e0f\u3093\u3067\u3001\u7740\u5b9f\u306b\u30b9\u30ad\u30eb\u3092\u8eab\u306b\u3064\u3051\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n
\n
- Python\u306e\u57fa\u790e\u6587\u6cd5\u3068\u30c7\u30fc\u30bf\u51e6\u7406\u30e9\u30a4\u30d6\u30e9\u30ea\uff08NumPy\u3001Pandas\uff09\u306e\u4f7f\u3044\u65b9\u3092\u7fd2\u5f97\u3059\u308b<\/li>\n\n\n\n
- \u6a5f\u68b0\u5b66\u7fd2\u306e\u57fa\u672c\u7684\u306a\u6982\u5ff5\uff08\u6559\u5e2b\u3042\u308a\u5b66\u7fd2\u3001\u640d\u5931\u95a2\u6570\u3001\u30aa\u30d7\u30c6\u30a3\u30de\u30a4\u30b6\u306a\u3069\uff09\u3092\u5b66\u3076<\/li>\n\n\n\n
- Keras\u306e\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3084\u66f8\u7c4d\u3001\u30aa\u30f3\u30e9\u30a4\u30f3\u8b1b\u5ea7\u3092\u6d3b\u7528\u3057\u3001Sequential\u30e2\u30c7\u30eb\u3068functional API\u306e\u4f7f\u3044\u65b9\u3092\u8eab\u306b\u3064\u3051\u308b<\/li>\n\n\n\n
- \u5b9f\u969b\u306e\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u4f7f\u3063\u3066\u3001\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u3068\u8a13\u7df4\u3092\u7e70\u308a\u8fd4\u3057\u7df4\u7fd2\u3059\u308b<\/li>\n\n\n\n
- CNN\u3001RNN\u3001Transformer\u306a\u3069\u306e\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306b\u3064\u3044\u3066\u5b66\u3076<\/li>\n\n\n\n
- \u30c8\u30e9\u30f3\u30b9\u30d5\u30a1\u30fc\u30e9\u30fc\u30cb\u30f3\u30b0\u3084\u8efd\u91cf\u30e2\u30c7\u30eb\u306a\u3069\u3001\u767a\u5c55\u7684\u306a\u30c8\u30d4\u30c3\u30af\u306b\u3082\u5f90\u3005\u306b\u53d6\u308a\u7d44\u3080<\/li>\n<\/ol>\n\n\n\n
\u5b9f\u8df5\u7684\u306a\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u30b9\u30ad\u30eb\u3092\u8eab\u306b\u3064\u3051\u308b\u306b\u306f\u3001\u624b\u3092\u52d5\u304b\u3057\u3066\u5b66\u3076\u3053\u3068\u304c\u5927\u5207\u3067\u3059\u3002\u4ee5\u4e0b\u306f\u3001MNIST\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u4f7f\u3063\u305f\u624b\u66f8\u304d\u6570\u5b57\u8a8d\u8b58\u30e2\u30c7\u30eb\u306e\u8a13\u7df4\u4f8b\u3067\u3059\u3002<\/p>\n\n\n\n
\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08MNIST\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u4f7f\u3063\u305f\u624b\u66f8\u304d\u6570\u5b57\u8a8d\u8b58\u30e2\u30c7\u30eb\u306e\u8a13\u7df4\uff09:<\/p>\n\n\n\n
from keras.datasets import mnist\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D\nfrom keras.utils import to_categorical\n\n(x_train, y_train), (x_test, y_test) = mnist.load_data()\nx_train = x_train.reshape((60000, 28, 28, 1)) \/ 255.0\nx_test = x_test.reshape((10000, 28, 28, 1)) \/ 255.0\ny_train = to_categorical(y_train)\ny_test = to_categorical(y_test)\n\nmodel = Sequential()\nmodel.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(28, 28, 1)))\nmodel.add(Conv2D(64, (3, 3), activation='relu'))\nmodel.add(MaxPooling2D(pool_size=(2, 2)))\nmodel.add(Dropout(0.25))\nmodel.add(Flatten())\nmodel.add(Dense(128, activation='relu'))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(10, activation='softmax'))\n\nmodel.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\nmodel.fit(x_train, y_train, batch_size=128, epochs=10, validation_data=(x_test, y_test))<\/pre>\n\n\n\n\u3053\u306e\u3088\u3046\u306a\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3092\u624b\u304c\u304b\u308a\u306b\u3001\u5b9f\u969b\u306bKeras\u3092\u4f7f\u3063\u3066\u307f\u308b\u3053\u3068\u304c\u4e0a\u9054\u3078\u306e\u8fd1\u9053\u3067\u3059\u3002<\/p>\n\n\n\n
Keras\u306f\u3001\u6a5f\u68b0\u5b66\u7fd2\u306e\u4e16\u754c\u3067\u6d3b\u8e8d\u3059\u308b\u305f\u3081\u306e\u5f37\u529b\u306a\u6b66\u5668\u3068\u306a\u308b\u3067\u3057\u3087\u3046\u3002\u521d\u5fc3\u8005\u306e\u6bb5\u968e\u304b\u3089\u3001\u7740\u5b9f\u306b\u30b9\u30ad\u30eb\u3092\u7a4d\u307f\u4e0a\u3052\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002\u65e5\u3005\u306e\u5b66\u7fd2\u306e\u7a4d\u307f\u91cd\u306d\u304c\u3001Keras\u3092\u4f7f\u3044\u3053\u306a\u305b\u308b\u30a8\u30f3\u30b8\u30cb\u30a2\u3078\u306e\u9053\u3092\u5207\u308a\u62d3\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
Keras\u3067\u5dee\u3092\u3064\u3051\u308b\u305f\u3081\u306b\u3001\u4eca\u65e5\u304b\u3089\u5b66\u7fd2\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u3066\u307f\u307e\u305b\u3093\u304b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u3053\u3093\u306b\u3061\u306f\u3002python\u3092\u4f7f\u3063\u305f\u6a5f\u68b0\u5b66\u7fd2\u306b\u8208\u5473\u304c\u3042\u308b\u3051\u308c\u3069\u3001\u306a\u304b\u306a\u304b\u4e00\u6b69\u304c\u8e0f\u307f\u51fa\u305b\u306a\u3044\u3068\u3044\u3046\u65b9\u306f\u3044\u307e\u305b\u3093\u304b\uff1f\u305d\u3093\u306a\u3042\u306a\u305f\u306b\u304a\u3059\u3059\u3081\u3057\u305f\u3044\u306e\u304c\u3001Keras\u3067\u3059\u3002Keras\u306f\u3001\u30b7\u30f3\u30d7\u30eb\u3067\u4f7f\u3044\u3084\u3059\u304f\u3001\u304b\u3064\u5f37\u529b\u306a\u6a5f\u68b0\u5b66\u7fd2\u30d5\u30ec\u30fc\u30e0 … <\/p>\n","protected":false},"author":1,"featured_media":640,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[],"class_list":{"0":"post-639","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-python"},"_links":{"self":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts\/639","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=639"}],"version-history":[{"count":1,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts\/639\/revisions"}],"predecessor-version":[{"id":641,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts\/639\/revisions\/641"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/media\/640"}],"wp:attachment":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=639"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=639"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=639"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}