import numpy  as np
 import pandas as pd
 from keras.preprocessing import  sequence
 from keras.models import  Sequential
 from keras.layers import  Dense, Dropout, Embedding, LSTM, Bidirectional
 from keras.datasets import  imdb
 from sklearn.metrics import  accuracy_score,classification_report
 
 # Max features are limited
 max_features = 15000  # 设置最大特征为15000
 max_len = 300 #单个句子最大长度为300
 batch_size = 64#每个批次大小
 
 # 加载数据
 (x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features)
 print(len(x_train),  'train observations')
 print(len(x_test),  'test observations')
 # 使得数据长度保持一致
 x_train_2 = sequence.pad_sequences(x_train, maxlen=max_len)
 x_test_2 = sequence.pad_sequences(x_test, maxlen=max_len)
 print('x_train  shape:', x_train_2.shape)
 print('x_test  shape:', x_test_2.shape)
 y_train = np.array(y_train)
 y_test = np.array(y_test)
 # 构建模型
 model = Sequential()
 model.add(Embedding(max_features, 128, input_length=max_len))
 model.add(LSTM(64))
 model.add(Dense(1, activation='sigmoid'))
 model.compile('adam', 'binary_crossentropy',  metrics=['accuracy'])
 # 训练模型
 model.fit(x_train_2, y_train,batch_size=batch_size,epochs=4,validation_split=0.2)
 # 模型预测
 y_train_predclass = model.predict_classes(x_train_2,batch_size=100)
 y_test_predclass = model.predict_classes(x_test_2,batch_size=100)
 y_train_predclass.shape = y_train.shape
 y_test_predclass.shape = y_test.shape
 # 输出模型评估结果
 print (("\nLSTM  Bidirectional Sentiment Classification   - Test accuracy:"),(round(accuracy_score(y_test,y_test_predclass),3)))
 print  ("\nLSTM  Bidirectional Sentiment Classification of Test data\n",classification_report(y_test,  y_test_predclass))
 print  ("\nLSTM  Bidirectional Sentiment Classification - Test Confusion Matrix\n\n",pd.crosstab(y_test,  y_test_predclass,rownames = ["Actuall"],colnames = ["Predicted"]))