Fabian Pedregosa

@zht012323 there hasn't been any progress on this

Although for SAG(A) this would be difficult because of the lagged updates implementation.

done for SAG in #42, still needs to be done for SVRG

Hi @casotto ! Thanks for the report. The solution sounds good. If you want you can make a pull request with the fix and I will re-generate the .cpp files.

Add tol=-1 to the kwargs of SAG*Classifier :-) Also, note that lightning has a definition the logistic loss that is very slightly different from that of liblinear. In particular, the...

Mathieu did this, but I would say speed as underflow can be controlled by other ways (liblinear does this pretty well).

Haven't seen such example, and didn't know about the technique. Feel free to submit an example if you think it could be useful On Jun 15, 2016 4:00 PM, "Levi...

yes we are looking at the residuals of the KKT conditions, "normalized" by the residuals at the first iteration if I remember correctly. Anyway, would be cool to have the...

thanks for the heads up. Don't hesitate to create a pull request with unfinished code if you want feedback before its finished

looking into it