Alexis Gidiotis

I have used this code extensively and in various datasets. My main observation is that the behaviour of the model mostly depends on the nature of the training data-target pairs....

You should call the function utils.visualize_attention anywhere in your test code. Another way to do it would be to add a secondary output to the attention layer that returns the...

I think that since you are getting output you are using the model correctly. The issue here is that the training hasn't converged. CTC is usually tricky to optimise so...

An update: I was trying to implement this myself and looks like this is not possible currently because the objects cannot be pickled. This is now related to issue #32

Game is still crashing on startup. System is Ubuntu 16.04 with NVIDIA 384.111 on vulkan. GL is able to start but I haven't checked the performance yet.

@spartian yes you absolutely can do this but it was not in the scope of this project.

Hi @spartian. Hierarchical is loosely based on [this paper ](https://www.cs.cmu.edu/~hovy/papers/16HLT-hierarchical-attention-networks.pdf). For more details check this [repo](https://github.com/AlexGidiotis/Document-Classifier-LSTM).