TypeError: Failed to convert object of type <class 'list'> to Tensor. Contents: [1, Dimension(None)]. Consider casting elements to a supported type.

[thushv89]

TypeError: Failed to convert object of type to Tensor. Contents: [1, Dimension(None)]. Consider casting elements to a supported type.

from attention_keras.

[habom2310]

I also got that problem, I change the shape in the Input and not use None
encoder_inputs = Input(shape=(None, n_inputs)) to encoder_inputs = Input(shape=(max_enc_seq_length, n_inputs)

[MaukWM]

I was getting the same error. I fixed this by changing my imports from

from keras.layers import Input, GRU, Dense, Concatenate, TimeDistributed from keras.models import Model

to

from tensorflow.python.keras.layers import Input, GRU, Dense, Concatenate, TimeDistributed from tensorflow.python.keras.models import Model

The issue seems to come from the fact that the keras layers do not give the full context of the shape to the Attention layer. Your model could not convert the list to a Tensor because there was a None type within (which is supposed to be a part of a shape).

[MaukWM]

I took another look and the solution @habom2310 proposes should fix the problem.

[JiweiZh]

I was getting the same error. I fixed this by changing my imports from

from keras.layers import Input, GRU, Dense, Concatenate, TimeDistributed from keras.models import Model

to

from tensorflow.python.keras.layers import Input, GRU, Dense, Concatenate, TimeDistributed from tensorflow.python.keras.models import Model

The issue seems to come from the fact that the keras layers do not give the full context of the shape to the Attention layer. Your model could not convert the list to a Tensor because there was a None type within (which is supposed to be a part of a shape).

@MaukWM
I changed all my imports by using tensorflow.python.keras.layers but I got the same error.......

[JiweiZh]

@MaukWM @habom2310 thx guys, that works!

[thushv89]

@MaukWM , @habom2310 and @Redtea0 ,

Apologies about the delay and lack of following up.

Yes, as some of you pointed out. You can only leave batch_size as None. All the other dimensions (i.e. sequence lenth and n_inputs needs to be fixed. This is because there are several reshaping steps performed during attention computation. And having more than one dimension set to None, it is impossible to compute the original shape from the reshaped input.

Hope that clears things up.

[amanagr]

Using both solutions together worked for me.

[theAB316]

As mentioned by @amanagr, using both solutions worked for me.

1. Adding a constant value instead on None
2. Changing the Keras related imports like,from keras.layers import to from tensorflow.keras.layers import

[Tam678]

@thushv89 ,hello,i met the same error.I try to solve the problem via solutions above.But how to add a constant value instead demension(3)?