Fwiw, we're using pylearn2 and blocks at Ersatz Labs. Both use Theano. I'd recommend them, particularly if you are into python. It's hard to build a good NN fr

Fwiw, we're using pylearn2 and blocks at Ersatz Labs. Both use Theano. I'd recommend them, particularly if you are into python. It's hard to build a good NN framework: subtle math bugs can creep in, the field is changing quickly, and there are varied opinions on implementation details (some more valid than others). Even just learning a single framework requires a good deal of effort.