Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Add newton_cg solver to TweedieRegression #16635

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
rth opened this issue Mar 4, 2020 · 4 comments
Closed

Add newton_cg solver to TweedieRegression #16635

rth opened this issue Mar 4, 2020 · 4 comments
Labels
module:linear_model New Feature

Comments

@rth
Copy link
Member

rth commented Mar 4, 2020 •
edited
Loading

The newton_cg solver for GLMs and TweedieRegression in particular was proposed in #9405 , it might be good to add it.

Preliminary benchmarks for it were done in #9405 (comment) .

If we add it, it would be good to re-use _newton_cg function from sklearn/utils/optimize.py as much as possible (also related #14248).

Also see: #16634 #16637
This was referenced Mar 4, 2020
Closed
Open
Closed
@ogrisel
Copy link
Member

ogrisel commented Mar 4, 2020

Prior to adding a new solver such as newton CG it would be good to understand when this solver is expected to out-perform the existing solver (L-BFGS). The linked preliminary benchmarks of the Tweedie regression PR could never highlight a case where newton CG would be competitive with either IRLS or L-BFGS.

@rth
Copy link
Member Author

rth commented Mar 4, 2020 •
edited
Loading

Prior to adding a new solver such as newton CG it would be good to understand when this solver is expected to out-perform the existing solver (L-BFGS)

Indeed, that is the main question. Currently newton_cg is only used in LogisticRegresion, and the UG is not very explicit as to when one would use it.

Thoughts about whether it would be useful to add @agramfort @lorentzenchr ?

@agramfort
Copy link
Member

agramfort commented Mar 4, 2020 via email

@lorentzenchr lorentzenchr mentioned this issue Mar 4, 2020
Closed
@rth
Copy link
Member Author

rth commented Mar 12, 2020

Closing as there is doesn't seem to be much motivation for this particular solver. Unless @lorentzenchr you object.

Would it make sense to deprecate newton-cg in LogisticRegression then given that we want to simplify its implementation #11865? also cc @NicolasHug

@rth rth closed this as completed Mar 12, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
module:linear_model New Feature
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@ogrisel @agramfort @rth