Self-Attention and other extensions

[loganbruns]

@ogroth, as mentioned in #27 I've found it useful to add self-attention and also some additional training options to limit the amount of data used for training. Also a simple test script. Would you please take a look at my branch and give me an idea of what parts you might consider including and I'll create a pull request with that subset?

master...loganbruns:sa_gqn

Thanks,
logan

[ogroth]

The self-attention encoder you've added in gqn_encoder.py looks neat. Can you send me a pull request for that on the attention_encoder branch?

[loganbruns]

@ogroth, thanks. I created a pull request with just the self-attention encoder in this pull request:

#30

I was wondering if you might be interested in a separate pull request against master with the limited training flags, --train_subset, from the branch above as well?