werwer

network.py

@@ -1,45 +1,57 @@
-from keras.preprocessing import sequence
-from keras.models import Sequential
-from keras.layers import Dense, Embedding
-from keras.layers import LSTM
-from keras.datasets import imdb
-
-max_features = 20000
-# cut texts after this number of words (among top max_features most common words)
-maxlen = 80
-batch_size = 32
-
-print('Loading data...')
-(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features)
-print(len(x_train), 'train sequences')
-print(len(x_test), 'test sequences')
-
-print('Pad sequences (samples x time)')
-x_train = sequence.pad_sequences(x_train, maxlen=maxlen)
-x_test = sequence.pad_sequences(x_test, maxlen=maxlen)
-print('x_train shape:', x_train.shape)
-print('x_test shape:', x_test.shape)
-
-
-print('Build model...')
-model = Sequential()
-model.add(Embedding(max_features, 128))
-model.add(LSTM(128, dropout=0.2, recurrent_dropout=0.2))
-model.add(Dense(1, activation='sigmoid'))
-
-# try using different optimizers and different optimizer configs
-model.compile(loss='binary_crossentropy',
-              optimizer='adam',
+import pandas as pd
+from sortedcontainers import SortedSet
+import numpy as np
+from sklearn.model_selection import train_test_split
+from keras.layers import Dense, Embedding, Input, Flatten
+from keras.layers import LSTM, GRU, Dropout
+from keras.models import Model
+import keras
+from keras.utils import plot_model
+import utils
+import time
+
+
+def build_model(want_answer_size, infact_answer_size):
+    inputs_want_answer = Input(shape=(want_answer_size, ), name='want_answer_input')
+    inputs_infact_answer = Input(shape=(infact_answer_size, ), name='infact_answer_input')
+    x_1 = Embedding(want_answer_size, 128, name='want_answer_embedding', embeddings_initializer='he_normal', embeddings_regularizer=keras.regularizers.l2(0.01))(inputs_want_answer)
+    x_2 = Embedding(infact_answer_size, 128, name='infact_answer_embedding', embeddings_initializer='he_normal', embeddings_regularizer=keras.regularizers.l2(0.01))(inputs_infact_answer)
+    x_1 = GRU(128, dropout=0.4, return_sequences=True, recurrent_initializer='he_normal', recurrent_regularizer=keras.regularizers.l2(0.01))(x_1)
+    x_2 = GRU(128, dropout=0.4, return_sequences=True, recurrent_initializer='he_normal', recurrent_regularizer=keras.regularizers.l2(0.01))(x_2)
+    x = keras.layers.concatenate([x_1, x_2])
+    x = Flatten()(x)
+    x = Dropout(0.3)(x)
+    x = Dense(64, activation='relu')(x)
+    predictions = Dense(2, activation='softmax')(x)
+    model = Model(inputs=[inputs_want_answer, inputs_infact_answer], outputs=predictions)
+    return model
+
+
+if __name__ == '__main__':
+    df = pd.read_excel('./预期输出与实际输出数据表.xlsx')
+    want_answer_corpus, infact_answer_corpus = utils.build_corpus(df)
+    onehot = utils.label2onehot(df['是否正确'])
+    x_train_1, x_test_1, y_train, y_test = train_test_split(want_answer_corpus, onehot, random_state=2333)
+    x_train_2, x_test_2, _, _ = train_test_split(infact_answer_corpus, onehot, random_state=2333)
+
+    want_answer_corpus_size = len(want_answer_corpus[0])
+    infact_answer_corpus_size = len(infact_answer_corpus[0])
+
+    model = build_model(want_answer_corpus_size, infact_answer_corpus_size)
+
+    # plot_model(model, to_file='model.png')
+
+    model.compile(loss='categorical_crossentropy',
+                  optimizer=keras.optimizers.Adam(lr=1e-4),
                   metrics=['accuracy'])
 
-print(model.summary())
-
-print('Train...')
-model.fit(x_train, y_train,
-          batch_size=batch_size,
-          epochs=15,
-          validation_data=(x_test, y_test))
-score, acc = model.evaluate(x_test, y_test,
-                            batch_size=batch_size)
-print('Test score:', score)
-print('Test accuracy:', acc)
+    print('Train...')
+    model.fit([x_train_1, x_train_2], y_train,
+              batch_size=16,
+              epochs=50)
+
+    score, acc = model.evaluate([x_test_1, x_test_2], y_test,
+                                batch_size=8, verbose=0)
+
+    print('Test score:', score)
+    print('Test accuracy:', acc)

utils.py

@@ -114,38 +114,7 @@ def label2onehot(label):
 
 
 if __name__ == '__main__':
-    df = pd.read_excel('./预期输出与实际输出数据表.xlsx')
-    want_answer_corpus, infact_answer_corpus = build_corpus(df)
-    onehot = label2onehot(df['是否正确'])
-    x_train_1, x_test_1, y_train, y_test = train_test_split(want_answer_corpus, onehot, random_state=2333)
-    x_train_2, x_test_2, _, _ = train_test_split(infact_answer_corpus, onehot, random_state=2333)
-
-    inputs_want_answer = Input(shape=(len(want_answer_corpus[0]), ), name='want_answer_input')
-    inputs_infact_answer = Input(shape=(len(infact_answer_corpus[0]), ), name='infact_answer_input')
-    x_1 = Embedding(len(want_answer_corpus[0]), 64, name='want_answer_embedding')(inputs_want_answer)
-    x_2 = Embedding(len(infact_answer_corpus[0]), 64, name='infact_answer_embedding')(inputs_infact_answer)
-    x_1 = GRU(64, dropout=0.5, return_sequences=0.2)(x_1)
-    x_2 = GRU(64, dropout=0.5, return_sequences=0.2)(x_2)
-    x = keras.layers.concatenate([x_1, x_2])
-    x = Flatten()(x)
-    x = Dense(64, activation='relu')(x)
-    predictions = Dense(2, activation='softmax')(x)
-    model = Model(inputs=[inputs_want_answer, inputs_infact_answer], outputs=predictions)
-
-    # plot_model(model, to_file='model.png')
-
-    model.compile(loss='categorical_crossentropy',
-                  optimizer='adam',
-                  metrics=['accuracy'])
-
-    print('Train...')
-    model.fit([x_train_1, x_train_2], y_train,
-              batch_size=16,
-              epochs=60)
-    score, acc = model.evaluate([x_test_1, x_test_2], y_test,
-                                batch_size=8)
-    print('Test score:', score)
-    print('Test accuracy:', acc)
+    pass