Docstring update for L2 penalty in SparseLogisticRegression #281
Conversation
This reverts commit 4563f22.
Docstring update for ElasticNet in SparseLogisticRegression (completes scikit-learn-contrib#244)
|
Hi @floriankozikowski you should avoid sending your PRs from your main branch, this will make your life difficult in the long run as your main and upstream/main will diverge For this PR it's Ok, but afterwards you should delete your local |
|
The linter fails because you have trailing whitespaces. Add |
skglm/estimators.py
Outdated
| .. math:: | ||
| \frac{1}{n_{\text{samples}}} \sum_{i=1}^{n_{\text{samples}}} | ||
| \log\left(1 + \exp(-y_i x_i^T w)\right) | ||
| + \alpha \cdot \left( \text{l1_ratio} \cdot \|w\|_1 + |
There was a problem hiding this comment.
go without the cdots for lighter formulas; also the squared L2 term should be divided by 2 (check the Enet docstring)
mathurinm
changed the title
|
The pytest failure is due to R install not to this PR : LGTM, merging |
Context of the PR
This PR finalizes and replaces #244, which was stalled. The PR adds ElasticNet regularization support to skglm.SparseLogisticRegression via the penalty="l1_plus_l2" option and l1_ratio parameter. All technical steps have been solved in previous PRs. This is just a docstring update.
Contributions of the PR
Updated class-level docstring and l1_ratio parameter doc to SparseLogisticRegression
Checks before merging PR