Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Conflicts:
BatchOptimization.lua
nnx-1.0-1.rockspec
|
|
|
|
|
|
|
|
|
|
|
|
It's basically a Sequential, where you can define the internal type
of computations, e.g. Cuda, Float or Double.
|
|
|
|
|
|
The goal is simple: instead of writing tons of Spatial*** modules,
we make sure all modules (Linear,SoftMax,...) can handle batches,
and use SpatialClassifier to replicate these modules spatially.
|
|
|
|
|
|
|
|
Conflicts:
OnlineTrainer.lua
|
|
ClassNLLCriterion only supports simple distributions, e.g. one-of-N.
DistNLLCriterion supports arbitrary distributions.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
(the reason is that those modules are going to be available
in the new torch version)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|