PyBrain library in parallel achitecture

Question

Is there someone who knows if pybrain works in parallel architecture? If it doesn't, which library is avaliable to work with Artificial Neural Network in multi-core processor? Thanks a lot!

Answer 1

You could try http://deeplearning.net/software/pylearn2/ -- it's written using Theano and so can run on multicore or GPU.

Answer 2

What you've asked depends on the task, but scikit-optimize is able to run in parallel using the ask-and-tell interface.

P = <parameter space>
optimizer = Optimizer(P)

while True:
    X = optimizer.ask(num_args=8)
    # X now has eight permutations of the parameter space
    Y = wait_for_parallel_evaluate(objective, X)
    # Y now has a list of results for those permutations
    optimizer.tell(X, Y)
    # optimizer uses the results to decide on what parameters it should try next

Instead of having this dedicated loop, you can also serialize optimizer using its dump function and bring it back to call tell on every parallel result and serialize it back.