End2EndModel.py 文件源码

def creat_binary_tag_LSTM( sourcevocabsize,targetvocabsize, source_W,input_seq_lenth ,output_seq_lenth ,
    hidden_dim ,emd_dim,loss='categorical_crossentropy',optimizer = 'rmsprop'):
    encoder_a = Sequential()
    encoder_b = Sequential()
    encoder_c = Sequential()
    l_A_embedding = Embedding(input_dim=sourcevocabsize+1,
                        output_dim=emd_dim,
                        input_length=input_seq_lenth,
                        mask_zero=True,
                        weights=[source_W])
    encoder_a.add(l_A_embedding)
    encoder_a.add(Dropout(0.3))
    encoder_b.add(l_A_embedding)
    encoder_b.add(Dropout(0.3))
    encoder_c.add(l_A_embedding)

    Model = Sequential()

    encoder_a.add(LSTM(hidden_dim,return_sequences=True))
    encoder_b.add(LSTM(hidden_dim,return_sequences=True,go_backwards=True))
    encoder_rb = Sequential()
    encoder_rb.add(ReverseLayer2(encoder_b))
    encoder_ab=Merge(( encoder_a,encoder_rb),mode='concat')
    Model.add(encoder_ab)

    decodelayer=LSTMDecoder_tag(hidden_dim=hidden_dim, output_dim=hidden_dim
                                         , input_length=input_seq_lenth,
                                        output_length=output_seq_lenth,
                                        state_input=False,
                                         return_sequences=True)
    Model.add(decodelayer)
    Model.add(TimeDistributedDense(targetvocabsize+1))
    Model.add(Activation('softmax'))
    Model.compile(loss=loss, optimizer=optimizer)
    return Model