在Enocder-Decoder架构中添加自定义注意层以使用keras进行神经机器翻译

问题描述

我正在尝试使用tensorflow-2 Keras将神经机器翻译成印地文的神经机器翻译任务。我已经创建了自己的自定义注意层，但是我无法弄清楚如何在编码器解码器层之间插入。而且我还需要一些帮助来准备用于训练和推理的数据集，因为那里有很多不同的实现方式，而且我感到困惑，所以如果有人可以帮助我，那将真的很有帮助！ 这是代码：

english_encoded_text = english_tokenizer.texts_to_sequences(english_data)
hindi_encoded_text = hindi_tokenizer.texts_to_sequences(hindi_data)

在这里，我已将数据转换为整数，现在我很困惑是填充这些序列还是一键填充！ 现在是模型：

#attention layer
class attention(Layer):
    def __init__(self,**kwargs):
        super(attention,self).__init__(**kwargs)

    def build(self,input_shape):
        self.W = self.add_weight(name="attention_weight",shape=(input_shape[-1],1),initializer="normal")
        self.b = self.add_weight(name="attention_bias",shape=(input_shape[1],initializer="zeros")        
        super(attention,self).build(input_shape)

    def call(self,x):
        energies = K.squeeze(K.tanh(K.dot(x,self.W) + self.b),axis=-1)
        alphas = K.softmax(energies)
        alphas = K.expand_dims(alphas,axis=-1)
        context_vector = x*alphas
        return K.sum(context_vector,axis=1)

    def compute_output_shape(self,input_shape):
        return (input_shape[0],input_shape[-1])

    def get_config(self):
        return super(attention,self).get_config()

#encoder-decoder with attention
latent_dim = 100
embedding_features = 256

s0 = Input(shape=(latent_dim,),name='s0') #initial hidden state for decoder
c0 = Input(shape=(latent_dim,name='c0') #initial cell state

encoder_inputs = Input(shape=(max_eng_sentence_length,english_vocab))
encoder_embedding =  Embedding(english_vocab,embedding_features,input_length=max_eng_sentence_length)(encoder_inputs)
encoder_lstm = Bidirectional(LSTM(latent_dim,return_sequences=True,dropout=0.3,recurrent_dropout=0.2))(encoder_embedding)
attention_context_vector = attention()(encoder_lstm)
decoder_layer = LSTM(latent_dim,return_state = True)
s,_,c = decoder_layer(attention_context_vector,initial_state=[s0,c0])
output_layer = Dense(hindi_vocab,activation="softmax")
out = output_layer(s)

model = Model(inputs=[X,s0,c0],outputs=out)

我遇到以下错误：

---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)

<ipython-input-21-1ea2aa5b4365> in <module>()
      8 encoder_inputs = Input(shape=(max_eng_sentence_length,english_vocab))
      9 encoder_embedding =  Embedding(english_vocab,input_length=max_eng_sentence_length)(encoder_inputs)
---> 10 encoder_lstm = Bidirectional(LSTM(latent_dim,recurrent_dropout=0.2))(encoder_embedding)
     11 attention_context_vector = attention()(encoder_lstm)
     12 decoder_layer = LSTM(latent_dim,return_state = True)

3 frames

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/layers/wrappers.py in __call__(self,inputs,initial_state,constants,**kwargs)
    528 
    529     if initial_state is None and constants is None:
--> 530       return super(Bidirectional,self).__call__(inputs,**kwargs)
    531 
    532     # Applies the same workaround as in `RNN.__call__`

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self,*args,**kwargs)
    924     if _in_functional_construction_mode(self,args,kwargs,input_list):
    925       return self._functional_construction_call(inputs,--> 926                                                 input_list)
    927 
    928     # Maintains info about the `Layer.call` stack.

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py in _functional_construction_call(self,input_list)
   1090       # Todo(reedwm): We should assert input compatibility after the inputs
   1091       # are casted,not before.
-> 1092       input_spec.assert_input_compatibility(self.input_spec,self.name)
   1093       graph = backend.get_graph()
   1094       # Use `self._name_scope()` to avoid auto-incrementing the name.

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/input_spec.py in assert_input_compatibility(input_spec,layer_name)
    178                          'expected ndim=' + str(spec.ndim) + ',found ndim=' +
    179                          str(ndim) + '. Full shape received: ' +
--> 180                          str(x.shape.as_list()))
    181     if spec.max_ndim is not None:
    182       ndim = x.shape.ndims

ValueError: Input 0 of layer bidirectional is incompatible with the layer: expected ndim=3,found ndim=4. Full shape received: [None,50,28103,256]

我需要帮助才能完成这项工作！ 预先感谢！

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）