Hello all,
Please review my proposal on improving the online learning for linear
models:
first draft - linear model proposal
<https://docs.google.com/document/d/1mRj-yxxhtapKGqFdA6M6e7PG6-klm0-1VJGBnFMUuU0/edit?usp=sharing>
Please bear in mind that this is a first approach only. I would like
your opinion on if this goes into the right direction and how it can be
improved.
On the tool to set the learning rate in particular, I need your ideas on
how it could be implemented. Previously mentioned ideas on a callback
function are interesting, but I would need some guidance on implementing
that.
Although I am interested in the decision trees as well, I feel the
linear model is a better start for me as my intention is to keep
contributing to scikit-learn after the summer. I have a background in
computational physics, however I am much more focused on the
computational side than the physics side. Here is my resume
<https://drive.google.com/file/d/0BwZy58HBIWp7U0I1RDNyRGhsNUU/view?usp=sharing>
.
PRs and Issues I have been involved in so far:
[MRG] enhance make_blobs to accept lists for samples per cluster
<https://github.com/scikit-learn/scikit-learn/pull/8563>
[MRG] add random_state in tests estimators
<https://github.com/scikit-learn/scikit-learn/pull/8563>
Bug in bfgs gradient computation of MLPRegressor with multiple output
neurons <https://github.com/scikit-learn/scikit-learn/issues/8349> (I am
very curious about this one)
github profile: kkatrio <https://github.com/kkatrio>
I am looking forward to your opinion.
Kind regards,
Konstantinos Katrioplas
_______________________________________________
scikit-learn mailing list
scikit-learn@python.org
https://mail.python.org/mailman/listinfo/scikit-learn
