It would be nice to generate some statistics about the dataset. E.g.

• distribution of genres
• synopsys lengths (considering that we define our constant MAX_SYNOPSIS_LEN)
• (maybe using a python notebook)
  Feel free to post some ideas

Now that we have some trained weights, we should code the functions necessary to load them into the model and make actual predictions. Goal is to be able to check whether the generated synopsis start to make sense or not.

This code should be useful for doing so

So far the network is not using any word embeddings like word2vec or GloVe. It could be interesting to compare results of that baseline against a net that makes use of them. I haven't figured out how to make them, basically because I used a lot of the code from this repo and he doesn't use it, although in the README.md says that Pre-trained word embeddings can also be used..

This is the best spanish pretrained word embedding model that I have found, so I think we should use them.

This should be useful https://gist.github.com/udibr/67be473cf053d8c38730

Use only top N frequent words to build vocabulary. We shouldn't take into account words that appear only a couple of times.

Train a network and use it as baseline for further improvements. This code should be useful