Parallel Pylearn2 layers

[gngdb]

It's possible to run Pylearn2 layers in parallel, and if we can do that then we can do create some much more interesting architectures, like pushing different preprocessing down different convolutional pipelines and merging in the fully connected layers. Also, we'll be able to put dummy layers in to push OpenCV features into the MLP layers.

[gngdb]

So far, found this pylearn-users on the topic. Seems like a good way to investigate this would be to do the job of the YAML parser and build a model up interactively; seeing which parts go where.

[gngdb]

I've been able to do this in a notebook Interactive Pylearn2 which is my notes on how to make a Pylearn2 model with parallel branches. Appears to work, although there are some extremely concerning Theano errors I've not dealt with yet.