CNN-LSTM模型

运行环境：python3.6.5 、Keras 2.1.5 、tensorflow 2.3.1等

CNN-LSTM的Sequential()写法：

from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.layers import Convolution1D, MaxPooling1D
from keras.layers import LSTM
#模型参数
time_step=100
# Convolution  卷积
filter_length = 5    # 滤波器长度
nb_filter = 64       # 滤波器个数
pool_length = 4      # 池化长度
# LSTM
lstm_output_size = 70   # LSTM 层输出尺寸
# Training   训练参数
batch_size = 30   # 批数据量大小
nb_epoch = 2      # 迭代次数
# 构建模型
model = Sequential()
model.add(Input(shape=(time_step, 128))  # 输入特征接收维度)  # 词嵌入层
model.add(Dropout(0.25))       # Dropout层
# 1D 卷积层，对词嵌入层输出做卷积操作
model.add(Convolution1D(nb_filter=nb_filter,
                        filter_length=filter_length,
                        border_mode='valid',
                        activation='relu',
                        subsample_length=1))
# 池化层
model.add(MaxPooling1D(pool_length=pool_length))
# LSTM 循环层
model.add(LSTM(lstm_output_size))
# 全连接层，只有一个神经元，输入是否为正面情感值
model.add(Dense(1))
model.add(Activation('sigmoid'))  # sigmoid判断情感（此处来做文本的情感分类问题）
model.summary()   # 模型概述
model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])
# 训练
model.fit(X_train, y_train, batch_size=batch_size, nb_epoch=nb_epoch,
          validation_data=(X_test, y_test))

输入[100, 128]的序列（100为序列长度，50为词嵌入维度），经过（池化层为MaxPooling1D，） Convlution1D(nb_filters=64, filter_length=5) 后，变为 [96, 64]，再经过 MaxPooling1D(pool_length=4) 后，变成了 [24, 64]。其模型维度结构如图所示：

Layer (type)                     Output Shape          Param #     Connected to
====================================================================================================
embedding_1 (Embedding)          (None, 100, 128)      2560000     embedding_input_1[0][0]
____________________________________________________________________________________________________
dropout_1 (Dropout)              (None, 100, 128)      0           embedding_1[0][0]
____________________________________________________________________________________________________
convolution1d_1 (Convolution1D)  (None, 96, 64)        41024       dropout_1[0][0]
____________________________________________________________________________________________________
maxpooling1d_1 (MaxPooling1D)    (None, 24, 64)        0           convolution1d_1[0][0]
____________________________________________________________________________________________________
lstm_1 (LSTM)                    (None, 70)            37800       maxpooling1d_1[0][0]
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 1)             71          lstm_1[0][0]
____________________________________________________________________________________________________
activation_1 (Activation)        (None, 1)             0           dense_1[0][0]
====================================================================================================

CNN-LSTM的结构化写法：

inputs = Input(shape=(1, 3))  # 输入特征接收维度
    a=Dropout(0.25)(inputs)
    conv=Convolution1D(10, 1,strides=1,padding="valid", dilation_rate=1)(a)#filters, kernel_size, strides=1
    max=MaxPooling1D(pool_length=pool_length)(conv)
    lstm1=LSTM(lstm_output_size)(max)
    output = Dense(1, activation='linear')(lstm1)  # 输出类别（此次来做简单的线性预测问题）
    model = Model(inputs=inputs, outputs=output)  # 初始命名训练的模型为model
    model.summary()

其模型维度结构如图所示：

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         (None, 1, 3)              0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 1, 3)              0         
_________________________________________________________________
conv1d_1 (Conv1D)            (None, 1, 10)             40        
_________________________________________________________________
max_pooling1d_1 (MaxPooling1 (None, 1, 10)             0         
_________________________________________________________________
lstm_1 (LSTM)                (None, 70)                22680     
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 71        
=================================================================
Total params: 22,791
Trainable params: 22,791
Non-trainable params: 0
_________________________________________________________________