search

lukesonnet / KRLS

R package for machine learning technique to fit flexible, interpretable functional forms for continuous and binary outcomes.
Other
21 stars 12 forks source link

Change lastkeeper to max at P for linear kernel #13

Closed lukesonnet closed 7 years ago

lukesonnet commented 7 years ago

Set this in numvectors

lukesonnet commented 7 years ago

Should we just warn if lastkeeper > P, and advise they set to P?

chadhazlett commented 7 years ago

rank(XX') should be at most P anyway so setting hard cap sound be fine.

On Sun, Aug 6, 2017 at 7:22 AM, Luke Sonnet notifications@github.com wrote:

Should we just warn if lastkeeper > P, and advise they set to P?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/lukesonnet/KRLS/issues/13#issuecomment-320510014, or mute the thread https://github.com/notifications/unsubscribe-auth/AFVCyJTmZHADezSJEH-X-uIi3v5C4jj3ks5sVcwpgaJpZM4OuqM8 .

lukesonnet commented 7 years ago

Okay, this currently sets max numvectors to try to P in the linear case and nrow(X) in all other cases. If for some reason it can't capture enough variance, it errors out and says that this should never happen but you can increase epsilon if you want.