https:\/\/aiandstory.net\/sample.zip<\/a> <\/p>\n\n\n\n# -*- coding: utf-8 -*-\n\nfrom __future__ import print_function\nfrom tensorflow.keras.models import Sequential, Model\nfrom tensorflow.keras.layers import Dense, LSTM, Embedding\nfrom tensorflow.keras.layers import Reshape, RepeatVector, TimeDistributed, Activation\nfrom tensorflow.keras.optimizers import Adamax\nfrom tensorflow.keras import Input\n\nimport numpy as np\nimport random\nimport sys\nimport warnings\nwarnings.filterwarnings('ignore')\n\n#utf-8\u306e\u30c6\u30ad\u30b9\u30c8\u3092\u6e96\u5099\npath=\"sample.txt\"\n\n#\u4e00\u62ec\u3067\u8aad\u307f\u8fbc\u307f\nwith open(path, encoding='utf-8') as f:\n text = f.read()\nprint('corpus length:', len(text))\n\n#\u7a7a\u767d\u3092\u9664\u53bb\ntokens = text.split()\ntext = ''.join(tokens)\n\n#\u6587\u5b57\u306e\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\u3092\u4f5c\u6210\u3059\u308b\u305f\u3081\u306b\u30bd\u30fc\u30c8\nchars = sorted(list(set(text)))\nprint('total chars:', len(chars))\n\n#\u6587\u5b57\uff1c\uff0d\uff1e\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\u306e\u8f9e\u66f8\u3092\u6e96\u5099\nchar_indices = dict((c, i) for i, c in enumerate(chars))\nindices_char = dict((i, c) for i, c in enumerate(chars))\n\n#\u5b66\u7fd2\u53ca\u3073\u6b63\u89e3\u30c7\u30fc\u30bf\u306e\u6e96\u5099\nmaxlen = 30#\u5165\u529b\u30c7\u30fc\u30bf\u306e\u6587\u5b57\u6570\u3002\u5143\u306e\u6587\u7ae0\u3092\u3053\u306e\u6587\u5b57\u6570\u3067\u5206\u5272\u3059\u308b\ngen_charlen = 1#\u51fa\u529b(\u6b63\u89e3)\u30c7\u30fc\u30bf\u306e\u6587\u5b57\u6570\nstep = 1\nsentences = []#\u5165\u529b\u30c7\u30fc\u30bf\u306e\u30ea\u30b9\u30c8\nnext_chars = []#\u51fa\u529b(\u6b63\u89e3)\u30c7\u30fc\u30bf\u306e\u30ea\u30b9\u30c8\nfor i in range(0, len(text) - maxlen - gen_charlen, step):\n sentences.append([char_indices[char] for char in text[i: i + maxlen]])\n next_chars.append([char_indices[char] for char in text[i + maxlen: i + maxlen + gen_charlen]])\nprint('input data sequences:', len(sentences))\n\n# build the model: \n\nprint('Build model...')\n\n#\u8a9e\u5f59\u6570\u304cEmbedding\u5c64\u306e\u5165\u529b\u6b21\u5143\u306b\u306a\u308b\nvcab_size = len(chars)\n\n#\u4ed6\u306e\u30d5\u30a1\u30a4\u30eb\u3067import\u3059\u308b\u5834\u5408\u3001\u4ee5\u964d\u306f\u5b9f\u884c\u3057\u306a\u3044\u3002\nif __name__ == \"__main__\":\n\n #\u6587\u5b57\u306e\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\u304c\u5165\u529b\u30c7\u30fc\u30bf\u306a\u306e\u3067int32\u3092\u6307\u5b9a,shape\u306b\u306f\u30b5\u30f3\u30d7\u30eb\u6570\u306e\u6b21\u5143\u306e\u8ef8\u306f\u542b\u307e\u306a\u3044\n #\u3053\u306e\u6307\u5b9a\u3067\u306fmaxlen\u6b21\u5143\u306e\u30d9\u30af\u30c8\u30eb\u304c\u5165\u529b\u3068\u306a\u308b\n text_input = Input(shape=(maxlen,), dtype='int32', name='text')\n #\u57cb\u3081\u8fbc\u307f\u5c64\u3002output_dim\u6b21\u5143\u306e\u30d9\u30af\u30c8\u30eb\u30b7\u30fc\u30b1\u30f3\u30b9\u306b\u57cb\u3081\u8fbc\u3080\n embedded_text = Embedding(input_dim=vcab_size ,output_dim=128)(text_input)\n \n hidden_unit = 128#LSTM\u306e\u96a0\u308c\u30e6\u30cb\u30c3\u30c8\u6570 \n output = embedded_text\n\n #RNN(LSTM)\u5c64\n output_enc = LSTM(hidden_unit, dropout=0.2, recurrent_dropout=0.5)(output)\n\n #\u51fa\u529b\u5c64\n output_dec = Dense(vcab_size, activation='softmax')(output_enc)\n\n #\u30e2\u30c7\u30eb\u306e\u30a4\u30f3\u30b9\u30bf\u30f3\u30b9\u5316\n model = Model(text_input, output_dec)\n \n optimizer = Adamax(lr=0.01)\n #softmax\u3067\u6570\u5024\uff08one-hot\u8868\u73fe\u3067\u306f\u306a\u304f\u6570\u5024\u306e\u914d\u5217\uff09\u306e\u30bf\u30fc\u30b2\u30c3\u30c8\u3092\u51e6\u7406\u3059\u308b\u5834\u5408\u306fspasparse_categorical_crossentropy\u3092\u6307\u5b9a\u3059\u308b\u3002\u3053\u306e\u6307\u5b9a\u306f\u3001\u5165\u529b\u30c7\u30fc\u30bf\u3054\u3068\u306b\u78ba\u7387\u30b9\u30b3\u30a2\u3092\u51fa\u529b\u3059\u308b\u305f\u3081\u3001\u30e2\u30c7\u30eb\u306e\u6700\u7d42\u7684\u306a\u51fa\u529b\u5f62\u72b6\u306b\u306f\u5f71\u97ff\u3057\u306a\u3044\n model.compile(loss='sparse_categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])\n model.summary()\n \n #model.fit()\u7528\u306bNumPy\u914d\u5217\u3092\u751f\u6210\u3059\u308b\u3002\n x = np.array(sentences)\n y = np.array(next_chars)\n\n #x\u306f\u5165\u529b\u30c7\u30fc\u30bf\u3001y\u306f\u51fa\u529b(\u6b63\u89e3)\u30c7\u30fc\u30bf\u3001epochs(\u8a66\u884c\u56de\u6570)\u306f\u4efb\u610f(10)\n model.fit(x, y,\n batch_size=128,\n epochs=10)\n\n #\u5b66\u7fd2\u6e08\u307f\u30e2\u30c7\u30eb\u3092\u4fdd\u5b58\n model.save('text_generation_for_aiandstory')\n <\/pre>\n\n\n\n\u53c2\u8003\u6587\u732e<\/strong>
https:\/\/github.com\/keras-team\/keras\/tree\/master\/examples<\/a> <\/p>\n\n\n\n<\/p>\n","protected":false},"excerpt":{"rendered":"
\u4efb\u610f\u306e\u9577\u3055\u30c6\u30ad\u30b9\u30c8\u3092RNN(LSTM)\u306b\u5b66\u7fd2\u3055\u305b\u308b\u305f\u3081\u306eKeras\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3002 \u5b66\u7fd2\u6e08\u307f\u306e\u30e2\u30c7\u30eb\u306f\u6587\u7ae0\u306e\u30a8\u30c3\u30bb\u30f3\u30b9\u3092\u8a18\u9332\u3057\u3066\u3044\u308b\u305f\u3081\u3001\u5165\u529b\u6587\u306b\u4f3c\u305f\u6587\u5b57\u5217\u3092\u51fa\u529b\u3059\u308b\u304c\u3001\u6587\u7ae0\u306e\u8cea\u306f\u826f\u304f\u306a\u3044\u3002 \u4e0b\u8a18\u306f\u30b3\u30fc\u30c9\u3067\u306f\u5b66\u7fd2\u53ca\u3073\u6b63 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[19,5],"tags":[11,13,20],"_links":{"self":[{"href":"https:\/\/aiandstory.net\/index.php?rest_route=\/wp\/v2\/posts\/315"}],"collection":[{"href":"https:\/\/aiandstory.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aiandstory.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aiandstory.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aiandstory.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=315"}],"version-history":[{"count":9,"href":"https:\/\/aiandstory.net\/index.php?rest_route=\/wp\/v2\/posts\/315\/revisions"}],"predecessor-version":[{"id":324,"href":"https:\/\/aiandstory.net\/index.php?rest_route=\/wp\/v2\/posts\/315\/revisions\/324"}],"wp:attachment":[{"href":"https:\/\/aiandstory.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=315"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aiandstory.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=315"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aiandstory.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=315"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}