Keras中间层（注意力模型）输出

数据挖掘 Python 喀拉斯 麻木的
2022-03-02 07:03:22

我有一个带有此摘要的模型：

___________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
input_1 (InputLayer)             (None, 30, 37)        0                                            
____________________________________________________________________________________________________
s0 (InputLayer)                  (None, 128)           0                                            
____________________________________________________________________________________________________
bidirectional_1 (Bidirectional)  (None, 30, 128)       52224       input_1[0][0]                    
____________________________________________________________________________________________________
repeat_vector_1 (RepeatVector)   (None, 30, 128)       0           s0[0][0]                         
                                                                   lstm_1[0][0]                     
                                                                   lstm_1[1][0]                     
                                                                   lstm_1[2][0]                     
                                                                   lstm_1[3][0]                     
                                                                   lstm_1[4][0]                     
                                                                   lstm_1[5][0]                     
                                                                   lstm_1[6][0]                     
                                                                   lstm_1[7][0]                     
                                                                   lstm_1[8][0]                     
____________________________________________________________________________________________________
concatenate_1 (Concatenate)      (None, 30, 256)       0           bidirectional_1[0][0]            
                                                                   repeat_vector_1[0][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[1][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[2][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[3][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[4][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[5][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[6][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[7][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[8][0]            
                                                                   bidirectional_1[0][0]            
                                                                   repeat_vector_1[9][0]            
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 30, 1)         257         concatenate_1[0][0]              
                                                                   concatenate_1[1][0]              
                                                                   concatenate_1[2][0]              
                                                                   concatenate_1[3][0]              
                                                                   concatenate_1[4][0]              
                                                                   concatenate_1[5][0]              
                                                                   concatenate_1[6][0]              
                                                                   concatenate_1[7][0]              
                                                                   concatenate_1[8][0]              
                                                                   concatenate_1[9][0]              
____________________________________________________________________________________________________
attention_weights (Activation)   (None, 30, 1)         0           dense_1[0][0]                    
                                                                   dense_1[1][0]                    
                                                                   dense_1[2][0]                    
                                                                   dense_1[3][0]                    
                                                                   dense_1[4][0]                    
                                                                   dense_1[5][0]                    
                                                                   dense_1[6][0]                    
                                                                   dense_1[7][0]                    
                                                                   dense_1[8][0]                    
                                                                   dense_1[9][0]                    
____________________________________________________________________________________________________
dot_1 (Dot)                      (None, 1, 128)        0           attention_weights[0][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[1][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[2][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[3][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[4][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[5][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[6][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[7][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[8][0]          
                                                                   bidirectional_1[0][0]            
                                                                   attention_weights[9][0]          
                                                                   bidirectional_1[0][0]            
____________________________________________________________________________________________________
c0 (InputLayer)                  (None, 128)           0                                            
____________________________________________________________________________________________________
lstm_1 (LSTM)                    [(None, 128), (None,  131584      dot_1[0][0]                      
                                                                   s0[0][0]                         
                                                                   c0[0][0]                         
                                                                   dot_1[1][0]                      
                                                                   lstm_1[0][0]                     
                                                                   lstm_1[0][2]                     
                                                                   dot_1[2][0]                      
                                                                   lstm_1[1][0]                     
                                                                   lstm_1[1][2]                     
                                                                   dot_1[3][0]                      
                                                                   lstm_1[2][0]                     
                                                                   lstm_1[2][2]                     
                                                                   dot_1[4][0]                      
                                                                   lstm_1[3][0]                     
                                                                   lstm_1[3][2]                     
                                                                   dot_1[5][0]                      
                                                                   lstm_1[4][0]                     
                                                                   lstm_1[4][2]                     
                                                                   dot_1[6][0]                      
                                                                   lstm_1[5][0]                     
                                                                   lstm_1[5][2]                     
                                                                   dot_1[7][0]                      
                                                                   lstm_1[6][0]                     
                                                                   lstm_1[6][2]                     
                                                                   dot_1[8][0]                      
                                                                   lstm_1[7][0]                     
                                                                   lstm_1[7][2]                     
                                                                   dot_1[9][0]                      
                                                                   lstm_1[8][0]                     
                                                                   lstm_1[8][2]                     
____________________________________________________________________________________________________
dense_2 (Dense)                  (None, 11)            1419        lstm_1[0][0]                     
                                                                   lstm_1[1][0]                     
                                                                   lstm_1[2][0]                     
                                                                   lstm_1[3][0]                     
                                                                   lstm_1[4][0]                     
                                                                   lstm_1[5][0]                     
                                                                   lstm_1[6][0]                     
                                                                   lstm_1[7][0]                     
                                                                   lstm_1[8][0]                     
                                                                   lstm_1[9][0]                     
====================================================================================================
Total params: 185,484
Trainable params: 185,484
Non-trainable params: 0
____________________________________________________________________________________________________

该模型进一步概括为：

在此处输入图像描述

而“注意”块总结为：

在此处输入图像描述

输入是模糊日期，例如“1979 年 11 月 17 日”（上限为 30 个字符），输出是 10 个字符表示“YYYY-mm-dd”。

我想绘制attention_weights图层的值。

我想看看网络在预测 、 和 中的每一个时“查看”“1979 年 11 月 17 日星期六”的YYYY哪个mm部分dd。我期待看到它完全忽略这一天（“星期六”）。

我尝试按照Keras 文档获取中间层的输出。

然而，注意力节点有 10 个输入，所以我必须抓住每一个：

f = K.function(model.inputs, [model.get_layer('attention_weights').get_output_at(t) for t in range(10)])
r = f([source, np.zeros((1,128)), np.zeros((1,128))])

例如“1979年source11 月 17 日”编码为

[[[ 0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]
  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
    1.]]]

r然后是一个形状矩阵(10,1,30,1)和我正在绘制的注意力图：

attention_map = np.zeros((10, 30))
for t in range(10):
    for t_prime in range(30):
        attention_map[t][t_prime] = r[t][0,t_prime,0]

...但是所有的值都是一样的！我期待一些变化。

我也试过添加K.learning_phase()无济于事。我究竟做错了什么？

1个回答

问题是我试图绘制从保存的模型加载的模型的注意力图。

保存模型时的输出为：

/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361：用户警告：lstm_1 层传递了不可序列化的关键字参数：{'initial_state': [, ]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。它们不会被包含在'/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361: UserWarning: Layer lstm_1 was passed non-serializable keyword arguments: {'initial_state': [，]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。
str（node.arguments）+'。它们不会被包含在'/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361: UserWarning: Layer lstm_1 was passed non-serializable keyword arguments: {'initial_state': [，]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。它们不会被包含在'/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361: UserWarning: Layer lstm_1 was passed non-serializable keyword arguments: {'initial_state': [，]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。它们不会包含在'/home/opyate/anaconda3/lib/python3. 6/site-packages/keras/engine/topology.py:2361：用户警告：lstm_1 层传递了不可序列化的关键字参数：{'initial_state': [, ]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。它们不会被包含在'/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361: UserWarning: Layer lstm_1 was passed non-serializable keyword arguments: {'initial_state': [，]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。它们不会被包含在'/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361: UserWarning: Layer lstm_1 was passed non-serializable key arguments: {'initial_state'：[，]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。它们不会被包含在'/home/opyate/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py:2361: UserWarning: Layer lstm_1 was passed non-serializable keyword arguments: {'initial_state': [，]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。他们不会被包括在内' 用户警告：层 lstm_1 传递了不可序列化的关键字参数：{'initial_state': [, ]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。他们不会被包括在内' 用户警告：层 lstm_1 传递了不可序列化的关键字参数：{'initial_state': [, ]}。它们不会包含在序列化模型中（因此在反序列化时会丢失）。str（node.arguments）+'。他们不会被包括在内'

但是，如果我从代码构建模型，并加载保存的权重，它就可以工作。

假设是UserWarning保存模型时的 s 与我的问题有关。

其它你可能感兴趣的问题
上一篇拼凑相当于 Google 的数据科学/工程 AIY 计算机可视化工具包 下一篇k- 表示马尔可夫链转移概率上的聚类