Hi Fred, I am also facing the same issue. Are you able to successfully implement this using the updated PR? Any idea if they have made changes to work with CPU also(as I don't have GPU)?
Thanks Asish On Thursday, February 11, 2016 at 6:39:43 AM UTC-5, Philipp Grothaus wrote: > > That's great. Thanks! I am using cudnn and will follow that PR. > > On Wednesday, February 10, 2016 at 7:28:31 PM UTC, nouiz wrote: >> >> There is a PR that allow this only on the cudnn version of pooling. So if >> you use cudnn and GPU, you can try this PR: >> >> Mostly, it is finished, excact that it break old pickled files. The PR >> author is working on this: >> >> https://github.com/Theano/Theano/pull/3965 >> >> Fred >> >> On Wed, Feb 10, 2016 at 8:43 AM, Philipp Grothaus <[email protected]> >> wrote: >> >>> Hi, >>> >>> I was wondering whether it is possible to keep the pooling size and >>> stride of signal.downsample.pool_2d unfixed at compile time. In other >>> words, is there a way to infer ds and st from the input dimensions through >>> a theano expression, instead of passing a tuple of integers to them? >>> >>> I'd like to do that in order to adapt the window size and stride to the >>> input dimensions of an image such that the output of the pooling operation >>> has a fixed dimension. >>> >>> Philipp >>> >>> -- >>> >>> --- >>> 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.
