Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
- renaming lsh -> lshift, rsh -> rshift
- adding componentwise functions
|
|
|
|
|
|
where appropriate with TH_REAL_IS_HALF), removed half from THGenerateAllTypes, added an explicit THGenerateHalfType.h
|
|
Add a new function bhistc to calculate histogram of batch of images only once
|
|
* Add support for torch.HalfTensor.
* Improvements/Simplifications for torch.HalfTensor.
Improvements/Simplifications:
1) Defines half type as TH_Half, so as to not conflict with cutorch
version. Previously, these were defined as the same "half" type and
required proper ordering of includes to ensure type was only defined
once, which would have affected all downstream projects.
2) No longer generates math functions that are not actually defined
on torch.HalfTensor, e.g. maskedFill, map, etc.
3) Adds tests for all available torch.HalfTensor functions
4) Allows compiling without TH_GENERIC_USE_HALF (so if there's a
problem can just unset that in CMakeLists rather than backing out)
5) Some simplifications: removes a new copy optimization and
some TH_HALF literal definitions
Limitations:
Because match functions are not defined, some "non-math" operators
on torch.HalfTensor give an error message, e.g. __index__/__newindex__
with a ByteTensor apply a mask, but masks aren't implemented. These
limitations aren't always obvious, (e.g. for documentation purposes),
but they should always give an error message.
* Rename TH_HALF to THHalf.
|
|
|
|
dimension, for speed reasons.
Fix a bug in cat when catting with an empty tensor along first dim (it added an extra dim).
Fix the ambiguous 'catting along last dimension' sentence in the doc and change the behavior to pick the maximum last dimension over all input tensors.
Now empty tensors are allowed.
|
|
|
|
|
|
|
|
Add fmod(), remainder(), remove mod() and fix tensor operator % behavior
|
|
Added torch.equal function which performs a tensor equality check
|
|
|
|
|
|
|
|
|
|
|
|
Add math functions trunc, frac, rsqrt, lerp
|
|
Fix torch.cat for table arguments
|
|
|
|
|
|
Now it also accepts a tensor of probabilities instead of a single Lua
number.
Added a test that checks that the output is binary.
|
|
|
|
|
|
specified dimension"
|
|
|
|
https://github.com/torch/torch7/issues/168
|
|
|
|
|
|
complete pivoting
|
|
|
|
add :neg() :csub(tensor) :csub(scalar), inplace subtraction operators
|
|
|
|
A function like `torch.mm` stores its output into a result tensor (res)
and uses the `addmm` function behind the scenes. `addmm` performs multi-
plications like res = beta x res + alpha x sum with beta=0 and alpha=1
here. If res is not initialized NaN values could occur resulting on NaN
output values.
The same applies for `torch.bmm` and `torch.mv`.
|
|
|
|
|
|
Exposing the lapack function potrs and extending torch.potrf() to take optional uplo character argument
|
|
adding tests for torch.potrs and (modified) torch.potrf
|
|
More flexible torch.cat
|
|
|
|
Adds a version of torch.cat that takes an array of tensors as inputs.
Adds TensorArray type to cwrap that reads an array of tensors from Lua
and creates a C array of tensor pointers that are passed to the C
function.
|
|
|
|
equations.
Fixed spaces in wrong place
|
|
Add cmin & cmax operation for element-wise min and max between two
tensors and between a tensor and a scalar.
|
|
lua_isstring returns true for a lua_Number. Then lua_tostring was called which led to a re-formatting (https://github.com/torch/torch7/issues/281) that generated a loss of precision for doubles who need more than 14 digits of precision.
Also avoids two useless conversions.
Replaced by the exclusive (lua_type(L, narg) == LUA_TSTRING) check.
|