将嵌入的嵌入与已有RNN中的整数输入相结合

我正在尝试在keras中创建一个状态的rnn，这是输入是嵌入和3个整数输入的串联。

以下是代码的简化版本，我添加了图层的尺寸作为注释，以使阅读

更容易

input = Input(batch_shape=(1,3,10))
inputTag = Lambda(lambda x: x[:,:,0:1])(input)
inputMeta =Lambda(lambda x: x[:,:,7:10])(input)
inputTag.shape, inputMeta.shape
#(TensorShape([Dimension(1), Dimension(3), Dimension(1)]), 
# TensorShape([Dimension(1), Dimension(3), Dimension(1), Dimension(3)]))
inputTagEnc = Embedding(tag_vocab_size,
                tag_emb_output_dim, 
                input_length = 1)(inputTag)
inputTagEnc.shape
#TensorShape([Dimension(1), Dimension(3), Dimension(1), Dimension(4)])
encodings =[inputTagEnc, inputMeta]
encodedInput = Concatenate()(encodings)
#Traceback (most recent call last):
#  File "<stdin>", line 1, in <module>
#  File "C:UsersmarcoAnaconda3libsite-packageskerasenginetopology.py", line 521, in __call__
#    self.build(input_shapes)
#  File "C:UsersmarcoAnaconda3libsite-packageskeraslayersmerge.py", line 153, in build
#    'Got inputs shapes: %s' % (input_shape))
#ValueError: `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(1, 1, 4), (1, 3, 1, 3)]

rnn = LSTM(n_hidden, 
             dropout=0.0, 
             activation='relu',
             recurrent_dropout=0.0,             
             stateful=True)(encodedInput)
output = Dense(3, activation='softmax')(rnn)

model = Model(inputs=[input], outputs=[encodedInput])
model.compile(loss='categorical_crossentropy', optimizer=Adam())
model.summary()

串联似乎正在失败，因为嵌入正在展开由于我的批量大小为3，但是整数输入正在尝试按原样处理。

现在，我尝试了一百万个重塑，变平和定时分配的变体，以使尺寸保持一致，但我仍然陷入困境。

有人可以解决这个问题吗？

class WorkingConcat(Layer):
def __init__(self, **kwargs):        
    super(WorkingConcat, self).__init__(**kwargs)
def build(self, input_shape):        
    super(WorkingConcat, self).build(input_shape)  # Be sure to call this somewhere!
def call(self, x):
    encodings, inputMeta = x[0:len(x)-1], x[len(x)-1]
    dd = K.reshape(inputMeta,(batch_size,3,1,3))
    encodings.append(dd)
    encodedInput = K.concatenate(encodings,axis=3)
    #ff = Reshape((3,67))(encodedInput)
    ff = K.reshape(encodedInput,(batch_size,3,67))
    self.output_dim = ff.shape
    return ff
def compute_output_shape(self, input_shape):
    return (batch_size, 3, 67)
encodings =[inputTagEnc]
encodings.append(inputMeta)
reshapedFrame = WorkingConcat()(encodings)