CUDA version of the train-on-cifar.lua

[srp1970]

Do we have CUDA version for this train-on-cifar.lua? I did the following:

1. Change as follows: -- load dataset torch.setdefaulttensortype("torch.CudaTensor") trainData = { data = torch.CudaTensor(50000, 3072), <== changed from torch.Tensor labels = torch.CudaTensor(50000),
2. Similar changes for "mlp" model case too.

But it gives the error saying:

preprocessing data (color space + normalization) /usr6/prakash/DNN/Torch/luajit-rocks/bin/luajit: invalid arguments: number number expected arguments: _CudaTensor_ [float] [float] stack traceback: [C]: at 0x7f2bb2949010 [C]: in function 'uniform' .../luajit-rocks/share/lua/5.1/nn/SpatialConvolutionMap.lua:103: in function 'reset' .../luajit-rocks/share/lua/5.1/nn/SpatialConvolutionMap.lua:73: in function '__init' ...kash/DNN/Torch/luajit-rocks/share/lua/5.1/torch/init.lua:54: in function <...kash/DNN/Torch/luajit-rocks/share/lua/5.1/torch/init.lua:50> [C]: in function 'SpatialConvolutionMap' ...cks/share/lua/5.1/nn/SpatialSubtractiveNormalization.lua:35: in function '__init' ...kash/DNN/Torch/luajit-rocks/share/lua/5.1/torch/init.lua:54: in function <...kash/DNN/Torch/luajit-rocks/share/lua/5.1/torch/init.lua:50> [C]: in function 'SpatialSubtractiveNormalization' ...cks/share/lua/5.1/nn/SpatialContrastiveNormalization.lua:23: in function '__init' ...kash/DNN/Torch/luajit-rocks/share/lua/5.1/torch/init.lua:54: in function <...kash/DNN/Torch/luajit-rocks/share/lua/5.1/torch/init.lua:50> [C]: in function 'SpatialContrastiveNormalization' train-on-cifar-cuda.lua:187: in main chunk [C]: in function 'dofile' ...Torch/luajit-rocks/lib/luarocks/rocks/trepl/scm-1/bin/th:131: in main chunk [C]: at 0x004051e0

[soumith]

have a look at https://github.com/szagoruyko/cifar.torch as well as a blog post here: torch.ch/blog/2015/07/30/cifar.html