Scalar Notes - gchanan/pytorch GitHub Wiki
General TODO
- generate scalar tests in test_autograd
- generate scalar tests in test_nn
- change binding so scalar matches tensor or scalar
- audit usage of size(dim), stride(dim) in derivatives.yaml
DONE TODO
- add requires_grad to python binding.
- add scalar tests for fill, index_fill, masked_fill
Axis functions
Reduction functions
| Function |
NumPy Empty tensor behavior |
Scalar tensor behavior |
Empty working? |
Scalar working? |
| min |
error |
identity |
|
|
| max |
error |
identity |
|
|
| kthvalue |
no equivalent, error |
identity |
|
|
| mode |
(SciPy) empty |
identity |
|
|
| median |
NaN |
identity |
|
|
| mean |
NaN |
identity |
|
|
| var |
NaN |
NaN |
|
|
| std |
NaN |
NaN |
|
|
| sum |
0 |
identity |
|
|
| prod |
1 |
identity |
|
|
| norm |
0 |
1 |
|
|
NN functions
| Function |
Empty tensor behavior |
Scalar tensor behavior |
Empty working? |
Scalar working? |
| glu |
Error |
Error |
|
|
| log_softmax |
|
|
|
|
| softmax |
|
|
|
|
Math functions
| Function |
Empty tensor behavior |
Scalar tensor behavior |
Empty working? |
Scalar working? |
| renorm |
|
|
|
|
| cumsum |
|
|
|
|
| cumprod |
|
|
|
|
| cross |
|
|
|
|
Tensor properties
| Function |
Empty tensor behavior |
Scalar tensor behavior |
Empty working? |
Scalar working? |
| size |
valid |
Error |
|
|
| stride |
valid |
Error |
|
|
Tensor shape
| Function |
Empty tensor behavior |
Scalar tensor behavior |
Empty working? |
Scalar working? |
| transpose |
|
|
|
|
| unfold |
|
|
|
|
| cat |
|
|
|
|
| chunk |
|
|
|
|
| permute |
|
|
|
|
| slice |
|
|
|
|
| split |
|
|
|
|
| stack |
|
|
|
|
| squeeze |
|
|
|
|
| unsqueeze |
|
|
|
|
Uncategorized functions
| Function |
Empty tensor behavior |
Scalar tensor behavior |
Empty working? |
Scalar working? |
| index_select |
|
|
|
|
| index_copy |
|
|
|
|
| index_add |
|
|
|
|
| index_fill |
|
|
|
|
| scatter_ |
|
|
|
|
| scatter_add |
|
|
|
|
| gather |
|
|
|
|
| sort |
|
|
|
|
| topk |
|
|
|
|
| select |
|
|
|
|