[scikit-learn] No convergence warning in logistic regression

[Andreas Mueller]

Having correlated data is not the same as not converging.
We could warn on correlated data but I don't think that's actually 
useful for scikit-learn.
I actually recently argued to remove the warning in linear discriminant 
analysis:
https://github.com/scikit-learn/scikit-learn/issues/14361

As argued in many places, we're not a stats library and as long as 
there's a well-defined solution,
there's no reason to warn.

LogisticRegression will give you the solution with minimum coefficient 
norm if there's multiple solutions.


On 9/2/19 5:40 AM, Guillaume Lemaître wrote:
> LBFGS will raise ConvergenceWarning for sure. You can check the 
> n_iter_ attribute to know if you really converged.
>
> On Mon, 2 Sep 2019 at 10:28, Benoît Presles 
> <benoit.presles at u-bourgogne.fr <mailto:benoit.presles at u-bourgogne.fr>> 
> wrote:
>
>     Hello Sebastian,
>
>     I have tried with the lbfgs solver and it does not change
>     anything. I do
>     not have any convergence warning.
>
>     Thanks for your help,
>     Ben
>
>
>     Le 30/08/2019 à 18:29, Sebastian Raschka a écrit :
>     > Hi Ben,
>     >
>     > I can recall seeing convergence warnings for scikit-learn's
>     logistic regression model on datasets in the past as well. Which
>     solver did you use for LogisticRegression in sklearn? If you
>     haven't done so, have used the lbfgs solver? I.e.,
>     >
>     > LogisticRegression(..., solver='lbfgs')?
>     >
>     > Best,
>     > Sebastian
>     >
>     >> On Aug 30, 2019, at 9:52 AM, Benoît Presles
>     <benoit.presles at u-bourgogne.fr
>     <mailto:benoit.presles at u-bourgogne.fr>> wrote:
>     >>
>     >> Dear all,
>     >>
>     >> I compared the logistic regression of statsmodels (Logit) with
>     the logistic regression of sklearn (LogisticRegression). As I do
>     not do regularization, I use the fit method with statsmodels and
>     set penalty='none' in sklearn. Most of the time, I have got the
>     same results between the two packages.
>     >>
>     >> However, when data are correlated, it is not the case. In fact,
>     I have got a very useful convergence warning with statsmodel
>     (ConvergenceWarning: Maximum Likelihood optimization failed to
>     converge) that I do not have with sklearn? Is it normal that I do
>     not have any convergence warning with sklearn even if I put
>     verbose=1? I guess sklearn did not converge either.
>     >>
>     >>
>     >> Thanks for your help,
>     >> Best regards,
>     >> Ben
>     >> _______________________________________________
>     >> scikit-learn mailing list
>     >> scikit-learn at python.org <mailto:scikit-learn at python.org>
>     >> https://mail.python.org/mailman/listinfo/scikit-learn
>     > _______________________________________________
>     > scikit-learn mailing list
>     > scikit-learn at python.org <mailto:scikit-learn at python.org>
>     > https://mail.python.org/mailman/listinfo/scikit-learn
>     _______________________________________________
>     scikit-learn mailing list
>     scikit-learn at python.org <mailto:scikit-learn at python.org>
>     https://mail.python.org/mailman/listinfo/scikit-learn
>
>
>
> -- 
> Guillaume Lemaitre
> INRIA Saclay - Parietal team
> Center for Data Science Paris-Saclay
> https://glemaitre.github.io/
>
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-learn/attachments/20190903/f99db952/attachment.html>