skglm

skglm copied to clipboard

REL - release 0.4

1

Requires: - [x] ~~#198~~ - [x] #197 - [x] #194 - [x] #137 - [x] #127 - [x] #160

mathurinm

FEAT - Add GroupLasso and GroupLogisticRegression estimators

14

mathurinm

## Context of the PR This PR adds support for intercept in `SqrtLasso` estimator. Closes #96 ### Checks before merging PR - [ ] added documentation for any new feature...

PascalCarrivain

DOC scrolling the left column menu scrolls the full page until bottom is reached, then scrolls the left menu

2

In the screencast my mouse is on the left column and I'm scrolling down [Screencast from 10-01-2024 09:11:36.webm](https://github.com/scikit-learn-contrib/skglm/assets/8993218/4759394b-bbae-44b3-9e7e-5f72a221038f)

mathurinm

FEAT - Use @njit(cache=True)

4

## Use caching option from Numba I explore a little bit the skglm source code and I realized you are using Numba decorator `@nijt`. I was wondering if it makes...

PascalCarrivain

FEAT - Add ``WeightedGroupL2_plus_L1`` penalty

1

## Context of the PR This PR adds a sparse weighted group L2 penalty enabling for instance fitting a sparse group Lasso estimator. Closes #198 ## Contributions of the PR...

Badr-MOUFAD

## Description of the feature Implementing fixed-point distance strategy to build working sets in ``GroupBCD`` will be very valuable. It enables solving problems with group penalties having a complex-to-evaluate subdiff...

Badr-MOUFAD

used in group lasso, group logreg subdifferential is a pain so we can skip it and support only fixpoint strategy

mathurinm

SCAD Regression Estimator

3

Implemented the SCAD regression class within estimators.py. This was inspired by the MCPRegression class in the skglm/estimators.py, and I wanted to create an analogous regression for the SCAD Penalty function.

v4gadkari

PR #43 is stale, retaking it over would be valuable namely for conciseness and readability of the ``MultiTaskBCD`` solver

Badr-MOUFAD