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.
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
RFC New parameters for penalties in LogisticRegression #28711
Comments
ping @scikit-learn/core-devs for more visibility. |
If we are to do such a big renaming, wouldn't it make sense to use a more explicit name such as Also note: the alpha in lasso / GLMs and the one in ridge do not have the same meaning. One is scaled by the sum of sample weights (or n_samples) while the other is not. If we are to use a more explicit name such as We had a discussion in the past about whether this is intentional or not because it could lead to make efficient parametrization for parameter tuning but I think we could decouple the choice of the default parameter value (which can stay estimator specific) from the choice of the parametrization which we could choose to make uniform across all linear models to simplify the message in the narrative doc, the reference doc and the various examples. We could also decide to move progressively and first rename C by |
But |
Indeed |
I would prefer a consistent name among all linear models and As of version 1.4:
|
Based on the comment #28706 (comment):
Currently,
LogisticRegression
usesC
as inverse penalization strength,penalty
to select the type of penalty andl1_ratio
to control the ration between l1 and l2 penalties.I propose the following:
alpha
(as inRidge
,ElasticNet
,PoissonRegressor
...) instead ofC
.Fail if both are given at the same time.
C
.penalty
which is redundant.alpha
andl1_ratio
are enough.The text was updated successfully, but these errors were encountered: