GitHub - EladWarshawsky/EfficientKANFormula: The original KAN implementation is slow and has a laborious formula conversion process, this aims to allow you to train KAN networks, at 10x the speed, while maintaining interpretability. Thanks to https://github.com/Blealtan/efficient-kan/tree/master

EladWarshawsky / EfficientKANFormula Public

Notifications You must be signed in to change notification settings
Fork 0
Star 0

The original KAN implementation is slow and has a laborious formula conversion process, this aims to allow you to train KAN networks, at 10x the speed, while maintaining interpretability. Thanks to https://github.com/Blealtan/efficient-kan/tree/master

0 stars 0 forks Branches Tags Activity

Star

Notifications

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
README		README
formula_solving.py		formula_solving.py
kan.py		kan.py
requirements.txt		requirements.txt
solving.ipynb		solving.ipynb

Repository files navigation

The original Efficient-KAN implementation has no formula conversion process, but is 10x faster than the original implementation.
This project is dedicate to being able to train Kolmogorov Arnold Networks quickly while still supporting interpretability.

Feel free to train a KAN on whatever task you like, then using the example script in solving.ipynb as a base, see if you can extract some interesting or informative formulas. 

TODO:
-Have more accurate formula approximation for the bsplines, it is not as good as it could be.