Forward AD rules and tests for polar decompositions#242
Conversation
|
🎉 CUDA tests are passing 🎉 |
Codecov Report❌ Patch coverage is
🚀 New features to boost your workflow:
|
lkdvos
left a comment
There was a problem hiding this comment.
Left some suggestions about performance but otherwise looks great!!
I wrote this in a less performant way but clearer to me way at the start just to make sure it was correct, thanks for the suggestions! Think I got everything in both push-forwards. |
lkdvos
left a comment
There was a problem hiding this comment.
It might even be worth it to add a comment to the implementations with the actual math for the formulas, you are right that it is way harder to follow what is going on now
|
We don't have that right now for other pullbacks so I am ok with this as-is, would be good to go through all of them at some point and add this though. I think the fail is unrelated, I'm going to merge. |
This seems to just work out of the box without hacking at special tangents or anything, so why not add...