koraykv / unsup

Some unsupervised learning modules using Torch

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

error when testing SpatialConvFistaL1

zhangguoxiao opened this issue · comments

Hi:
As I try to test SpatialConvFistaL1,I find that the code in line 59 is wrong.So I change the code as:
local conntable = nn.tables.full(params.nfiltersout, params.nfiltersin)
mlp = unsup.SpatialConvFistaL1(conntable, params.kernelsize, params.kernelsize, params.inputsize, params.inputsize, params.lambda)
data:conv()
The error message is :stack traceback:
[C]: in function 'copy'
/home/systex/torch/install/share/lua/5.1/unsup/FistaL1.lua:113: in function 'updateOutput'
demo_fista.lua:71: in function 'updateSample'
demo_fista.lua:82: in function 'train'
Finally I find out that the code in 113 in FistaL1.lua seem to be wrong.Am I right?or how should I make thing right!

I change the code in line 86 of file FistaL1.lua to :
if params.inputsize == params.kernelsize and params.conv == false then
serr, siter = updateSample(example[1], example[2] ,currentLearningRate)
else
icode=torch.Tensor(params.nfiltersout):zero()
serr, siter = updateSample(example[1], icode ,currentLearningRate)
end
It works!But I don't think it is a good solution!Any suggestions?