【问题标题】:How to display all LSTM states in Keras summary?如何在 Keras 摘要中显示所有 LSTM 状态?
【发布时间】:2019-07-12 09:33:31
【问题描述】:

当使用 keras 功能 model.summary() 时,它向我展示了我的模型的张量形状,非常棒! 不幸的是,当使用编码器 LSTM 时,使用带有属性 return_states=Truekeras.layers.LSTM 构造函数调用,摘要不会以其完整形式显示。它看起来像这样:

Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input (InputLayer)              (None, 34, 30)       0                                            
__________________________________________________________________________________________________
encoder (LSTM)                  [(None, 34, 30), (No 7320        input[0][0]                      
__________________________________________________________________________________________________
lambda_8 (Lambda)               (None, 34, 15)       0           encoder[0][0]                    
__________________________________________________________________________________________________
decoder (LSTM)                  (None, 34, 30)       5520        lambda_8[0][0]                   
                                                                 encoder[0][1]                    
                                                                 encoder[0][2]                    
==================================================================================================
Total params: 12,840
Trainable params: 12,840
Non-trainable params: 0
__________________________________________________________________________________________________

如您所见,编码器的输出形状被截断,只有三个形状中的第一个可见。有没有办法显示它,也许是修复甚至解决方法? :)

【问题讨论】:

    标签: keras deep-learning state lstm summary


    【解决方案1】:

    找到解决方法:

    print(encoder.output_shape)
    >> [(None, 34, 30), (None, 30), (None, 30)]
    

    【讨论】:

      猜你喜欢
      • 2017-08-31
      • 2018-04-04
      • 1970-01-01
      • 2018-03-30
      • 2018-03-24
      • 1970-01-01
      • 1970-01-01
      • 1970-01-01
      • 2017-10-08
      相关资源
      最近更新 更多