experiments/nmt default state

[babillu]

I observed the following minor issue.

This is what is written in readme for experiments/nmt:

Simply running ... train.py ... would start training in the current directory. ... The default prototype state used is _prototype_searchstate that corresponds to the RNNsearch-50 model from [1].

It turns out that by "simply running train.py" I get an error message, because train.py parse_args() contains: ("--proto", default="prototype_state")

[babillu]

experiments/nmt readme.md:

When training starts, look for the following files:

• search_model.npz contains all the parameters
• search_state.pkl contains the state
• search_timing.npz contains useful training statistics

In fact, file names: encdec_model.npz encdec_state.pkl encdec_timing.npz

[rizar]

That depends on the state prototype you use. If you used "prototype_searchstate", the file names would be "search*".