Changes Needed to Make Code Run at Notre Dame

[klannon]

This is following up on an email exchange I had with @peterjsadowski:

My student (@DayOfThePenguin) and I tried to run this code out of the box, but we ran into two problems:

• The code crashed due to exhausting the memory of the computer we had available. We were able to fix this by reorganizing how the how the numpy array was filled. We didn't do rigorous testing to confirm what was causing the problem, but I suspect that the main issue was that there were multiple statements of the form X = X[:,1:], which I believe causes python to make a copy of X behind the scenes for assignment. I think multiple instances of copies being made in combination with python garbage collection was leading to the memory exhaustion.
• Some of the constructor arguments and/or available pylearn2 objects seem to be different for the version of pylearn2 used for this code compared to what we have here at Notre Dame. We've installed pylearn2 from the git master branch as of the last month or so. Is it possible that this version of pylearn2 isn't compatible with the one used to develop this code?

Since we only "fixed" one of the files for running the training (and "fixed" is probably too loose of a term--we commented out lines causing crashes until the code ran), then I won't be hurt if you reject this PR. However, I thought it was the easiest way to share with you what changes we needed to make to get things up and running.