lmjohns3

add feed_forward-like method for computing loss value

1

We should include the value of the loss in the output from the feed_forward method.

Some operations are going to require 4D inputs and outputs (e.g., 3D convolutions). We should add models to support this, probably just a copy of the theanets/recurrent.py module ported to...

Recurrent models should accept nonzero initial state

1

To sample sequentially from a recurrent model, it would be useful to add some API for accepting an initial hidden state and running the model forward one time step.

add "bayesian rnn" implementation

1

http://arxiv.org/pdf/1512.05287v1.pdf Yarin Gal has a paper describing an LSTM that uses a "fixed" dropout mask to remove LSTM cells. It would be cool to have an implementation of this in...

We should try to make sure theanets doesn't leak!