Keras: 'TensorVariable' object has no attribute 'get_output_shape_at'

[JiaqingFu]

Please make sure that the boxes below are checked before you submit your issue. Thank you!

• [ ] Check that you are up-to-date with the master branch of Keras. You can update with:
  pip install git+git://github.com/fchollet/keras.git --upgrade --no-deps
• [ ] If running on Theano, check that you are up-to-date with the master branch of Theano. You can update with:
  pip install git+git://github.com/Theano/Theano.git --upgrade --no-deps
• [ ] Provide a link to a GitHub Gist of a Python script that can reproduce your issue (or just copy the script here if it is short).

[JiaqingFu]

I am implement a model ,that take input: a sequence of image and a sequence of word. And I do this
with keras functional API. the code is below:

VGG_model = VGG_16()
VGG_model.trainable = False
from keras.layers import Input, Dense
from keras.models import Model
input_sequences = Input(shape=(5,3,244,244))
processed_sequences = TimeDistributed(VGG_model)(input_sequences) #outputshape=(None, 5, 4096)
processed_sequences=GRU(output_dim=4096, return_sequences=False)(processed_sequences)
processed_sequences=RepeatVector(1)(processed_sequences)#(None, 1, 4096)

max_caption_len = 21
vocab_size = 43
input_sequences_0=Input(shape=(max_caption_len-1,), dtype='int32')
input_sequences_l=Embedding(input_dim=vocab_size, output_dim=256,input_length=max_caption_len-1)(input_sequences_0)
input_sequences_l=GRU(output_dim=128, return_sequences=True)(input_sequences_l)
input_sequences_l=TimeDistributed(Dense(4096))(input_sequences_l)

merged=Merge([processed_sequences,input_sequences_l],mode='concat', concat_axis=1)

model = Model(input=[input_sequences,input_sequences_0], output=merged)

But I get the error at "merged=Merge([processed_sequences,input_sequences_l],mode='concat', concat_axis=1)
"
the error message is below, I am new to keras, Is anyone know why? thanks in advance

AttributeError                            Traceback (most recent call last)
<ipython-input-5-e58a6d91d287> in <module>()
     15 input_sequences_l=TimeDistributed(Dense(4096))(input_sequences_l)#(None,max_caption_len-1,4096)input_sequences_lEGD=TimeDistributedDense(4096)(input_sequences_lEG)#(None,max_caption_len,4096)
     16 
---> 17 merged=Merge([processed_sequences,input_sequences_l],mode='concat', concat_axis=1)
     18 
     19 model = Model(input=[input_sequences,input_sequences_0], output=merged)

C:\Anaconda2\lib\site-packages\keras\engine\topology.pyc in __init__(self, layers, mode, concat_axis, dot_axes, output_shape, node_indices, tensor_indices, name)
   1112                                        concat_axis, dot_axes,
   1113                                        output_shape,
-> 1114                                        node_indices, tensor_indices)
   1115             self.built = True
   1116             self.add_inbound_node(layers, node_indices, tensor_indices)

C:\Anaconda2\lib\site-packages\keras\engine\topology.pyc in _arguments_validation(self, layers, mode, concat_axis, dot_axes, output_shape, node_indices, tensor_indices)
   1135         input_shapes = []
   1136         for i, layer in enumerate(layers):
-> 1137             layer_output_shape = layer.get_output_shape_at(node_indices[i])
   1138             if type(layer_output_shape) is list:
   1139                 # case: the layer has multiple output tensors

AttributeError: 'TensorVariable' object has no attribute 'get_output_shape_at'

[joelthchao]

Can you at least use syntax highlighting for your code?

[JiaqingFu]

@joelthchao I has edited it ,Is it clear now?

[joelthchao]

Merge is for model, while merge is for tensor. You can check the code and some useful docstrings.

from keras.layers import merge
merged = merge([processed_sequences, input_sequences_l], mode='concat', concat_axis=1)

[JiaqingFu]

@joelthchao
Thanks for help, now the error is gone !