search

KindXiaoming / pykan

Kolmogorov Arnold Networks
MIT License
14.94k stars 1.38k forks source link

Why is it that when I run hellokan, it runs so much differently than expected #258

Closed yIJunWangg closed 3 months ago

yIJunWangg commented 4 months ago

Here's the result of my run image

xwbxxx commented 4 months ago

I have the same problem in my issue #254. And I have 2 more problems. Do you meet the other 2 problems like I do?

Crusty-Banana commented 3 months ago

I think this happened because there is still some hidden degree of random in the code (between device). You can try increasing the training set or changing up the seed of the create_dataset() function

kingkongabc commented 3 months ago

Same here. Is it possible to be instability with floating point in the implementation of LBFGS?

KindXiaoming commented 3 months ago

Adding this line in the beginning helps

torch.use_deterministic_algorithms(True)
KindXiaoming commented 3 months ago

This should be solved with the most up-to-date version 0.2.0. Close this issue for now.

yIJunWangg commented 3 months ago

This should be solved with the most up-to-date version 0.2.0. Close this issue for now.

thank you