Clean-up reductions - Githubissues

Quansight-Labs / numpy_pytorch_interop

19 stars 4 forks source link

Clean-up reductions #131

Closed lezcano closed 1 year ago

lezcano commented 1 year ago

Avoid the repetition of the signature of reductions
Fix keepdims. With the previous implementation it was not possible to set it as a positional arg.
Fix some functions that were incorrectly exported in torch_np. Added a test for this as well.
- To go with this, I had to move the imports back into einsum to avoid having an from _normalizations import and a import _normalizations right after it, as discussed.

lezcano commented 1 year ago

I realized we are exporting a few more things than we need (decorators) in linalg and fft. Should we do something about those? We'd need to qualify their calls, which is a bit annoying.

ev-br commented 1 year ago

Should we do something about those? We'd need to qualify their calls, which is a bit annoying.

del them in the __init__.py?