Never mind, It's solved by using conv2d operation instead of directly subtensor operations! LOL
在 2017年4月19日星期三 UTC+8上午12:04:18，carter Frank写道： > > Hi, > I am using theano to define a loss function for the training of CNN, > like Loss = T.sqrt(Gx**2+Gy**2). > My prediction G is a M*N matrix and Gx/Gy are a horizontal/vertical > gradient of the matrix prediction G, they have the same size M*N. Gx[i,j] > = G[i+1,j]-G[i,j], Gy[i,j] = G[i,j+1]-G[i,j], > I quite new to theano, how can I solve this fastly, thanks! > > Best,Fed > -- --- 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.
