[scikit-learn] Interest in decision thresholding and fairness ? pr [10117]

[Prokopis Gryllos]

Dear sklearn folks,

Some time ago (almost two years now :D) I started working on this pr
https://github.com/scikit-learn/scikit-learn/pull/10117 ; The starting
point was an issue created by Andreas Mueller
https://github.com/scikit-learn/scikit-learn/issues/8614. Andreas and Joel
Nothman helped with reviewing all iterations up to a few months ago but for
the most part the pr has been stalled.

I'd like to ask the community if there is still interest for this to move
forward? I think in today's state of ML decision thresholding is important;
Day to day data science work frequently requires thinking about thresholds
and many times people settle for intuitive solutions or thresholds that are
not tested for generalisation. Furthermore using more advanced ways for
choosing thresholds is not trivial. The increased interest in fairness in
ML also points to the idea that thresholding is important.

I am not sure if this is the best place to ask but I wasn't sure how else I
could find out if people are interested in reviewing the work. I'd like to
put some effort into finalising the feature and potentially adapt it to the
current state of the art. Would something like that
http://papers.nips.cc/paper/6374-equality-of-opportunity-in-supervised-learning.pdf
be of interest?

Looking forward to your responses

Gr,
Prokopis
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-learn/attachments/20190828/de284911/attachment.html>