So, back to decorating the classes and not the objects so that torch.save() does...
authorFrancois Fleuret <francois@fleuret.org>
Tue, 6 Dec 2016 08:07:06 +0000 (09:07 +0100)
committerFrancois Fleuret <francois@fleuret.org>
Tue, 6 Dec 2016 08:07:06 +0000 (09:07 +0100)
commitb8c7166b9123735e8226d34b717d3cbc2dc1fa02
treef0bdf74bbb29444f041f212dae7036cdcee94add
parentc2a852c17eb19f472762343ee8dfe3e0d1b40dbf
So, back to decorating the classes and not the objects so that torch.save() does not complain with SpatialConvolution. Added the possibility to pass the total time to profiler.print() so that the fraction of time used by the different functions can be displayed.
profiler.lua
test-profiler.lua