search

Blealtan / efficient-kan

An efficient pure-PyTorch implementation of Kolmogorov-Arnold Network (KAN).
MIT License
3.92k stars 349 forks source link

Necessity of the scale parameter? #1

Closed KindXiaoming closed 5 months ago

KindXiaoming commented 5 months ago

Hi, I just want to share my experience when I developed KAN that the scale parameter seems quite important (but that was in the very beginning of the KAN project, so I could be hallucinating). Would love to hear your experimental results! Great initiative, would love to see a more efficient KAN implementation (with good features maintained).

Blealtan commented 5 months ago

Thank you for your information. It's added back now with a controlling option, but I haven't yet done any experiments on that.

KindXiaoming commented 5 months ago

[heart] Ziming Liu reacted to your message:

From: Blealtan @.> Sent: Friday, May 3, 2024 8:53:31 AM To: Blealtan/efficient-kan @.> Cc: Ziming Liu @.>; Author @.> Subject: Re: [Blealtan/efficient-kan] Necessity of the scale parameter? (Issue #1)

Thank you for your information. It's added back now with a controlling option, but I haven't yet done any experiments on that.

— Reply to this email directly, view it on GitHubhttps://github.com/Blealtan/efficient-kan/issues/1#issuecomment-2092584870, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AFTV5B2YZSM7RC32UZAU54DZANGAXAVCNFSM6AAAAABHEVQEVSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAOJSGU4DIOBXGA. You are receiving this because you authored the thread.Message ID: @.***>