You probably don't have the same problem. Start a new thread on the mailing list with a good description of the problem and your error.
Otherwise I'm pretty sure nobody will help you. Fred Le 20 janv. 2017 03:44, "Dewesh Pardhi" <dewesh...@gmail.com> a écrit : > Hi Pranav I am also facing same issue ,not able to compute the grad of one > of the layer,can you pl help me out,you can reach out to me at > dewesh...@gmail.com ,mo-+91 9096014001 <+91%2090960%2014001> > > On Sunday, August 28, 2016 at 4:54:15 PM UTC+5:30, pranav inani wrote: >> >> >> >> On Saturday, 27 August 2016 22:46:23 UTC+5:30, martin.de...@gmail.com >> wrote: >>> >>> From what you say you have 4 parameters if I understand correctly >>> [W,bW,b]? >>> >> >> W,b and W,b are the weights and biases of the hidden layer and logistic >> classifer respectively(the last two layers of a conv neural network). The >> next six values (<TensorType(float64, 4D)>, <TensorType(float64, vector)>, >> <TensorType(float64, 4D)>, <TensorType(float64, vector)>, >> <TensorType(float64, 4D)>, <TensorType(float64, vector)>) are supposed to >> be the weights and biases of the 3 ConvPool layers preceding the hidden >> layer. >> >> >> >>> Are they referring to the same thing? if not consider renaming them >>> W1,b1,W2,b2 >>> >>> From your code you define your parameters to be float32 but here from >>> what you say your params when you print them are float64. >>> Why is that? >>> >>> Explicitly define every parameter that you use to be float32, also make >>> sure your data are float32 for consistency. >>> >> >> I have coded everything to be in floatX taht defaults to float 64. The >> code that works (the one without the loop) works fine with float 64. I >> haven't changed the dtype parameters anywhere and it works in the other >> code. So I have concluded that it is not the problem. >> >> https://groups.google.com/forum/#!topic/theano-users/FQz7qn3C1qc >> >> >> >> above is a link to a similar problem. I have tried that as well. >> >> I tried feeding the params separately for each layer. Interestingly, the >> grad function was able to calculate the gradients for the hidden layer, >> logistic classifier and the 3rd ConvPool layer. But gave disconnected input >> error for the first 2 ConvPool layers. >> >> All this leads to one thing: the fault is in the loop. >> >> Any other ideas would be appreciated. >> >> Regards >> Pranav >> >> >> >> >>> >>> -- > > --- > 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 theano-users+unsubscr...@googlegroups.com. > 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 theano-users+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
