Hi Taemin, You've probably found this out by now but I just came across your question and thought I'd try and help. The real-valued numbers are interpreted as probabilities for the binary units and that's why they work for the normal rbm. Seems you can use the probabilities for the visible states, but you have to use binary for the hidden states.
Cheers, Jim On Thursday, 10 September 2015 09:30:08 UTC+1, Taemin Jo wrote: > > Hello. > > Sample DBN is pre-trained by RBM, not GRBM. > > RBM only can handle 0, 1(, theano_rng.binomial means that). > > DBN can learn sample MNIST well, what is more its inputs are real numbers. > > Thank you. > -- --- 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.
