Philippe Rémy

I'd say it's because we have a custom function to build() the network. The support for RaggedTensor should be about updating this function. That's my gut feeling.

Yes, I think you are right here. It's related. I'll see what I can do ;)

The correct way to do that is here: https://www.tensorflow.org/guide/keras/masking_and_padding Doc: https://stackoverflow.com/questions/55176818/how-to-support-masking-in-custom-tf-keras-layers-layer

Install tensorflow 1.15

Time to move to pytorch.

@kwon3969 we consider inputs with shape `(batch_size, time_steps, input_dim)`. The attention weights (after softmax) have shape `(batch_size, time_steps)`. The `input_dim` is no longer here. You can only know where the...

@AnoopSaju123 thank you!

@rgommers thanks. let's wait for the next pytorch version then!

One dep is missing in the conda requirements. https://anaconda.org/conda-forge/zbar

You can either install it at the OS level or in conda. ``` conda install -c conda-forge pyzbar ```