[scikit-learn] n_jobs for LogisticRegression

[Sebastian Raschka]

Thanks, Tom, that makes sense. Submitted a PR to fix that.

Best,
Sebastian

> On Dec 19, 2016, at 10:14 AM, Tom DLT <tom.duprelatour at orange.fr> wrote:
> 
> Hi,
> 
> In LogisticRegression, n_jobs is only used for one-vs-rest parallelization.
> In LogisticRegressionCV, n_jobs is used for both one-vs-rest and cross-validation parallelizations.
> 
> So in LogisticRegression with multi_class='multinomial', n_jobs should have no impact.
> 
> The docstring should probably be updated as you mentioned. PR welcome :)
> 
> Best,
> Tom
> 
> 2016-12-19 6:13 GMT+01:00 Sebastian Raschka <se.raschka at gmail.com>:
> Hi,
> 
> I just got confused what exactly n_jobs does for LogisticRegression. Always thought that it was used for one-vs-rest learning, fitting the models for binary classification in parallel. However, it also seem to do sth in the multinomial case (at least according to the verbose option). in the docstring it says
> 
> >     n_jobs : int, optional
> >         Number of CPU cores used during the cross-validation loop. If given
> >         a value of -1, all cores are used.
> 
> and I saw a logistic_regression_path being defined in the code. I am wondering, is this just a workaround for the LogisticRegressionCV, and should the n_jobs docstring in LogisticRegression
> be described as "Number of CPU cores used for model fitting” instead of “during cross-validation,” or am I getting this wrong?
> 
> Best,
> Sebastian
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn
> 
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn