I use theano.sandbox.rng_mrg.MRG_RandomStreams to generate a 2D (and 1D ) dropout masks for applying dropouts. I create it so,
srng = MRG_RandomStreams(rng.randint(1,89324723894), use_cuda = True) mask = srng.binomial (n=1, p=1-dropout_rate,size=params.shape,dtype = floatX) otuput = params *mask. doing this operation is doing two things. 1. If I use the libgpuarray backend, and I have two GPUS, the mask is being created in another GPU from the one I have used in config.device. 2. If I use it with use_cuda = False, it is creating the data in the same GPU, but then the code does not run on GPU.. Appreciate any help. -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
