Replies: 1 comment
-
Hey @arthus701 - I saw you have an example that you're working through with the KeOps folks (getkeops/keops#156). Let me know how this is going. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi everyone,
I am struggling with a memory efficient implementation of the kernels mentioned here and here. Conceptually, they are kernels for vector valued outputs, that include some linear transformation/rotation of the vector coordinate system. After the discussion over at #1453, I am trying to implement them as KeOpsKernels.
Does anyone have experience with pulling off something similar to this MultiOutput-Kernel with KeOps? Or is the current restriction to two "indexing" dimensions of KeOps permitting this? I guess you wouldn't need more symbolic dimensions, just a way to broadcast or extend the KeOps-tensors.
If it helps I could also provide the PyTorch implementation that I currently have.
Thank you,
Arthus
Beta Was this translation helpful? Give feedback.
All reactions