get_config

[philipperemy]

get_config

[philipperemy]

Which version of tensorflow are you using? Update your tensorflow.

[edithangelicar]

I have tensorflow 2.3.0.

For example, at the end of your example (example-attention.py ) after "model.fit" I wrote:

model.fit(x_train, y_train, validation_data=(x_val, y_val), epochs=max_epoch,
batch_size=64, callbacks=[VisualiseAttentionMap()])

#What I wrote
model.save('test_model.h5')
modelh5 = load_model('test_model.h5', custom_objects={'Attention': Attention})

Error I get:

File "C:\Users\edith\anaconda3\lib\site-packages\tensorflow\python\keras\layers\merge.py", line 663, in build
raise ValueError('A Dot layer should be called '

ValueError: A Dot layer should be called on a list of 2 inputs.

I am attaching a picture of the script where I try to save and load the model

Thank you very much for your help and time!

[philipperemy]

Good catch I think we need to use Dot instead of dot.

[philipperemy]

Hum still does not work and not sure why:

[philipperemy]

@edithangelicar save can sometimes be very tricky. I advise you to use load_weights and save_weights instead:

You will have to re-define your model and instanciate your model beforehand but I don't see any other solutions in the short term.

model.save_weights('test_model.h5')

model = <bla bla>
model.load_weights('test_model.h5')

[edithangelicar]

Thank you for checking and for your help!
If at some point you get to know how to solve the problem, please let me know.
Thank you!

[edithangelicar]

I think I managed, you have to change your model from sequential to functional API. Thank you!

[nickion]

Have literally just hit this issue myself; thanks for the comments and update.

[edithangelicar]

Nickion, also remember to create the get_config and from_config in the class: https://keras.io/guides/serialization_and_saving/

[philipperemy]

Great @edithangelicar can you provide a small snippet to show how to get it to work?

[edithangelicar]

yes. here are some pictures, it is lightly modified for fitting what I am doing so it doesn't correspond exactly to your example.

[philipperemy]

@edithangelicar can you paste it here instead of pictures? It will be easier for others to understand it.

[edithangelicar]

I am attaching txt since here I can't paste python files

functional_lstm.txt
attention.txt

[philipperemy]

Alright thanks!