Question

我正在尝试定义双向层的初始状态，我无法弄清楚如何为该层定义初始状态。我可以为简单的lstm RNN定义初始状态，但是当我将层更改为双向时，我不知道如何定义初始状态。

这是一个简单的例子

h_in = K.reshape(tf.constant([2.]),(1,1))
c_in = K.reshape(tf.constant([3.]),(1,1))
lstm = Bidirectional(LSTM(1, return_sequences=True),merge_mode='concat')
l_lstm = lstm(sequence_input , initial_state= [h_in, c_in, h_in, c_in])
l1 = Dense(1, activation = 'linear')(l_lstm)
model = Model(inputs = [sequence_input], outputs =  l1)

它返回以下错误：

-
TypeError                                 Traceback (most recent call last)
<ipython-input-93-d5838abcb4be> in <module>
     11 
     12 lstm = Bidirectional(LSTM(1, return_sequences=True),merge_mode='concat')
---> 13 l_lstm = lstm(sequence_input , initial_state=[h_in, c_in, h_in, c_in])
     14 # ( sequence_input , initial_state= init_state )
     15 # l0 = Dense(40, activation='softplus')(l_lstm)

~\Anaconda3\lib\site-packages\keras\layers\wrappers.py in __call__(self, inputs, initial_state, constants, **kwargs)
    481             return output
    482         else:
--> 483             return super(Bidirectional, self).__call__(inputs, **kwargs)
    484 
    485     def call(self,

~\Anaconda3\lib\site-packages\keras\engine\base_layer.py in __call__(self, inputs, **kwargs)
    455             # Actually call the layer,
    456             # collecting output(s), mask(s), and shape(s).
--> 457             output = self.call(inputs, **kwargs)
    458             output_mask = self.compute_mask(inputs, previous_mask)
    459 

~\Anaconda3\lib\site-packages\keras\layers\wrappers.py in call(self, inputs, mask, training, initial_state, constants)
    516                 backward_inputs += inputs[-self._num_constants:]
    517             y = self.forward_layer.call(forward_inputs,
--> 518                                         initial_state=forward_state, **kwargs)
    519             y_rev = self.backward_layer.call(backward_inputs,
    520                                              initial_state=backward_state, **kwargs)

~\Anaconda3\lib\site-packages\keras\layers\recurrent.py in call(self, inputs, mask, training, initial_state)
   2192                                       mask=mask,
   2193                                       training=training,
-> 2194                                       initial_state=initial_state)
   2195 
   2196     @property

~\Anaconda3\lib\site-packages\keras\layers\recurrent.py in call(self, inputs, mask, training, initial_state, constants)
    603             mask = mask[0]
    604 
--> 605         if len(initial_state) != len(self.states):
    606             raise ValueError('Layer has ' + str(len(self.states)) +
    607                              ' states but was passed ' +

TypeError: object of type 'Tensor' has no len()

为Keras中的双向层定义状态的初始值

0 个答案: