This is a know problem. We have prs that should fix that. I think you can just use them for now.
I didn't see any real benefit of ignore_border=False. If you find some, tell me. But cudnn don't support that, so it will always be faster to set it to True. There is an issue about this, but I'm offline, so I can't link to it. Fred Le 13 nov. 2016 00:40, "Ragav Venkatesan" <[email protected]> a écrit : > in theano.tensor.signal.pool.pool2d if I use ignore_border = False, > things get pulled off GPU. If I use ignore_border = True, things run on GPU > > Using code cnn_tutorial from the theano tutorials. > > -- > > --- > 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. > -- --- 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.
