model_rnn.py 文件源码

def bidirectional_gru(len_output):
    # sequence_input is a matrix of glove vectors (one for each input word)
    sequence_input = Input(
        shape=(MAX_SEQUENCE_LENGTH, EMBEDDING_DIM,), dtype='float32')
    l_lstm = Bidirectional(GRU(100))(sequence_input)
    # TODO look call(input_at_t, states_at_t) method, returning (output_at_t, states_at_t_plus_1)
    # also look at switch(condition, then_expression, else_expression) for deciding when to feed previous state
    preds = Dense(len_output, activation='softmax')(l_lstm)
    model = Model(sequence_input, preds)
    model.compile(loss='categorical_crossentropy',
                  optimizer='rmsprop',
                  metrics=[utils.f1_score, 'categorical_accuracy'])

    return model

# required, see values below