Age | Commit message (Collapse) | Author |
|
|
|
Added SpatialAdaptiveMaxPooling
|
|
|
|
Euclidean batch support
|
|
WeightedEuclidean optimizations
|
|
|
|
|
|
Added more informative pretty-printing.
|
|
|
|
CMul optimizations, doc and unit tests
|
|
|
|
new Copy constructor arguments
|
|
Parallel, Container & cie
|
|
|
|
|
|
|
|
|
|
|
|
|
|
refactoring all the common container code into nn.Container
|
|
|
|
speedup and optimizations for SparseLinear
|
|
|
|
another fix for making nn.Linear non-blocking
|
|
|
|
added optional precision parameter to nn.Jacobian
|
|
|
|
nn.Linear blocking fix
|
|
|
|
segfaults over multiple threads (on some compilers)
|
|
correction in batch mode
|
|
|
|
|
|
WeightedEuclidean batch mode
|
|
|
|
VolumetricConvolution batch mode + test
|
|
|
|
Temporary buffers for unfolding (for CUDA version)
|
|
|
|
CMul batch processing
|
|
Changed CMul reset of weights to random initialization
|
|
removing the requirement for providing size in nn.Mul
|
|
|
|
|
|
generic/LogSoftMax speedups
|
|
LookupTable + Concat small fixes
|
|
|
|
|
|
|
|
[Torch] Move test.lua to the top level
|