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

Skip to content

Add L1 penalty and coordinate descent solver to TweedieRegression #16637

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
Open
rth opened this issue Mar 4, 2020 · 3 comments
Open

Add L1 penalty and coordinate descent solver to TweedieRegression #16637

rth opened this issue Mar 4, 2020 · 3 comments
Labels
module:linear_model New Feature

Comments

@rth
Copy link
Member

rth commented Mar 4, 2020

The L1 penalty with coordinate descent solver for GLM was proposed in #9405 and was not included in the minimal implementation merged in #14300.

It would be be good to add it, however care must to taken to avoid redundancy with the existing coordinate descent solver in ElasticNet.

Preliminary benchmarks were done in #9405 (comment)

Also see: #16634 #16635
@rth rth added the New Feature label Mar 4, 2020
This was referenced Mar 4, 2020
Closed
Closed
Closed
@lorentzenchr
Copy link
Member

IIRC, one reason not to use the current enet_coordinate_descent solver for #9405 was that it does not naturally support sample weights nor an intercept term. The implementation in #9405 seems to use more memory than necessary. On the other side, given the pure python implementation, I'd say it was computationally acceptable.

@lorentzenchr lorentzenchr mentioned this issue Mar 4, 2020
Closed
@Reksbril
Copy link
Contributor

Reksbril commented Apr 28, 2020 •
edited
Loading

@rth can I work on this?

@lorentzenchr lorentzenchr mentioned this issue Nov 21, 2020
Open
@lorentzenchr lorentzenchr mentioned this issue Jan 27, 2022
Closed
@mathurinm
Copy link
Contributor

For users interested, in the meantime we have implement a Poisson datafit in the sklearn compatible skglm package, which can be used with a ProxNewton solver for example.

Such datafit can be combined with a wealth of penalties (L1, but also WeightedL1, Elastic net - L1 + L2 -, and non convex penalties such as MCP, SCAD, Lq penalties with q < 1).

We also have the Gamma datafit; feedback is welcome!

@glemaitre glemaitre moved this to Discussion in Losses and solvers May 17, 2024
@lorentzenchr lorentzenchr mentioned this issue May 16, 2025
Open
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
Losses and solvers
Status: Discussion
Milestone
No milestone
Development

No branches or pull requests
4 participants
@rth @mathurinm @lorentzenchr @Reksbril