'Embedding' object has no attribute 'get_shape'

[lordt...@gmail.com]

I was experimenting with `Embedding` layer, that I need for ignoring some values in sequences (masking).

The code is the following one:

import numpy as np
from keras.layers import Input, Dense, LSTM
from keras.layers.embeddings import Embedding
from keras.layers.merge import Concatenate
from keras.models import Model
from keras.utils import plot_model

trainExs = np.asarray([ [1, 2, 3], [2, 3, 1]])
trainLabels = np.asarray([[1, 1, 1], [2, 2, 2]])

print('Examples, shape:', trainExs.shape)
print('Labels, shape:', trainLabels.shape)

W = [[0, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]]
symDim = 3

# E M B E D D I N G S
symbol_emb = Embedding(symDim+1, symDim,
 weights=np.asarray(W), trainable=False, mask_zero=True)
symbol_dense = Dense(symDim, use_bias=True, name='symbol_dense')(symbol_emb)

output_layer = Dense(symDim, dtype='float32', name='output')(symbol_dense)

# M O D E L
model = Model(inputs=[symbol_emb], outputs=[output_layer])
model.compile(loss='mean_squared_error', optimizer='RMSprop', metrics=['accuracy'])
# print(model.summary())

However, when I run it, I get the following error messages:

Examples, shape: (2, 3)
Labels, shape: (2, 3)
Using TensorFlow backend.
Traceback (most recent call last):
 File "D:/workspace/TESTS/test/testEMb.py", line 26, in <module>
 symbol_dense = Dense(symDim, use_bias=True, name='symbol_dense')(symbol_emb)
 File "D:\python\lib\site-packages\keras\engine\topology.py", line 541, in __call__
 self.assert_input_compatibility(inputs)
 File "D:\python\lib\site-packages\keras\engine\topology.py", line 450, in assert_input_compatibility
 ndim = K.ndim(x)
 File "D:\python\lib\site-packages\keras\backend\tensorflow_backend.py", line 479, in ndim
 dims = x.get_shape()._dims
AttributeError: 'Embedding' object has no attribute 'get_shape

From the documentation I am not able to understand how to solve it.

I need the embedding for going from symbol indexes (in trainExs) to one-hot encodings, after concatenated by another a layer.

Is there anyone who can help me?