Conversation
|
Hello all, Just out of curiosity, did you consider the efficient KAN implementation? Honestly speaking, I did not managed to compare those two implementations, but if you see a performance-related issue, may be worth to do so. I can take a look, later. |
* Improve spline * Add KAN --------- Co-authored-by: Filippo Olivo <folivo@filippoolivo.com>
Add future todos on kan_block
|
Hi, I've fixed almost all the KAN-related issues. There are still several failing tests, but they are inherited from 0.3 at the moment. The |
|
Hi, a quick update on this PR:
Pending tasks before merge:
I plan to complete these items and finalize the PR by the end of next week. |
Description
The PR fixes the KAN model and block to be compatible with the latest
Splinemodel. Moreover, a vectorized spline model is also implemented for efficiency.I also added a test on the KAN training, and it is working. Despite that, the current model still needs (too) many iterations to converge. Further validation before merging 0.3 in master are needed.
This PR fixes #612
Checklist