implement warm restarts for logistic regression about scikit-learn HOT 12 CLOSED

@MechCoder what do you think about it?

from scikit-learn.

Sure, but let me finish this PR first.

from scikit-learn.

@arjoly @fabianp I believe this involves patching the liblinear files too, such that accept the coef_init as a parameter, am I right?

from scikit-learn.

@GaelVaroquaux Should I patch the liblinear files too so that there is uniformity between all the three classes?

from scikit-learn.

Sorry, this issue is 4 years old and now I have a different opinion.

The more we patch liblinear the harder it is to keep up with upstream.
Plus, now that we have different solvers based on Scipy.optimize we can do
warm restarts (which we take advantage of in LogisticRegressionCV).

My opinion would be to close this issue as obsolete thanks to Manoj's
work.
On Jul 28, 2014 1:13 PM, "Manoj Kumar" [email protected] wrote:

@arjoly https://github.com/arjoly @fabianp https://github.com/fabianp
I believe this involves patching the liblinear files too, such that accept
the coef_init as a parameter, am I right?

—
Reply to this email directly or view it on GitHub
#11 (comment)
.

from scikit-learn.

I understand, perhaps just adding it to the newton-cg and lbfgs solvers, for Logistic Regression would be a better option?

from scikit-learn.

Indeed

from scikit-learn.

Can we close this issue now that we have the new solvers?

from scikit-learn.

Can we close this issue now that we have the new solvers?

I think so. Good catch

from scikit-learn.

I just did ;)

from scikit-learn.

We could still have a warm start option for the new solvers though, but I'm not sure if it is needed.

from scikit-learn.

+1 for closing, thanks

from scikit-learn.