DOC - Specify penalty in GammaRegressor documentation
#24789
Conversation
sklearn/linear_model/_glm/glm.py
Outdated
|
|
||
| This regressor uses the 'log' link function. | ||
| Its optimization objective is | ||
|
|
||
| .. math:: |
There was a problem hiding this comment.
For refs, check:
GammaRegressor documentationGammaRegressor documentation
sklearn/linear_model/_glm/glm.py
Outdated
| Its optimization objective is:: | ||
|
|
||
| sum(Xw_i + y_i * exp(-Xw_i) - log(y_i) - 1) | ||
| + (1/2) * alpha ||w||_2^2 |
There was a problem hiding this comment.
I am still not sure that we want to have the formulation here.
@lorentzenchr What is your feeling about it? If we start to introduce the objective function here, then we probably want to do the same for the other GLM.
There was a problem hiding this comment.
I personally like the objective as part of the docstring, but we often only have them in the user guide, examples: LogisticRegression, PoissonRegressor, GammaRegressor.
In case we include them in the docstring, having them as
There was a problem hiding this comment.
Cons: It could be considered as implementation detail. It bloats the docstring. We would have it in 2 places, user guide and docstring.
Pros: Without objective, one does not know what the model does. Only the exact objective enables to compare with other libraries.
GammaRegressor documentationGammaRegressor documentation
|
I agree with you @glemaitre on #24789 (comment), better not to add the objective function expression in this place. I removed it and kept the details about the penalty term and the doc formatting. |
|
Thanks @Badr-MOUFAD LGTM. |
Reference Issues/PRs
Not linked to any exiting issue or PR.
What does this implement/fix? Explain your changes.
This fixes
GammaRegressordocumentation. In particular:Any other comments?
No.