Looking at the code, where is use_noise defined? Generally you will have a symbolic scalar which is set to 1 during training and 0 during test (or reversed, whatever), then in the tensor.switch you tensor.eq(use_noise,1) to determine what code path (dropout/no dropout) to use during that run. >From the SO code, you don't ever define use_noise as anything (you simply pass it to dropout_layer in _lstm_e, but it isn't defined as anything or passed into the function).
On Friday, September 23, 2016 at 11:29:33 AM UTC-4, Peter Potash wrote: > > I do pass the updates to the theano function (you can check the > stackoverflow link). I feel more and more that the problem is passing a > symbolic shape to the binomial rng. Others have had this issue, as I allude > to in my post. > > -Peter > -- --- 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.
