Le 18 avril 2012 13:54, Ian Goodfellow <[email protected]> a écrit :
> I'm not sure if it's correct. I'm also not familiar enough with the
> library to be sure that no similar functionality exists somewhere. (I
> think David Warde-Farley mentioned that there was a plan to do
> something like this at some point)
> Can anyone comment on whether this is actually the decision function
> that OneVsRestClassifier uses internally?

You can write a unittest on a multiclass setup (e.g. 3 classes) as a
sanity check, against LinearSVC:

Fit a OneVsRestClassifier(LinearSVC()) and a LinearSVC() directly on
the same training set and compare the outcome of the
`decistion_function()` call on those two models on the same test set.
As liblinear is already using one-vs-the-rest internally for the
multiclass case, this should yield the same results.

-- 
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel

------------------------------------------------------------------------------
Better than sec? Nothing is better than sec when it comes to
monitoring Big Data applications. Try Boundary one-second 
resolution app monitoring today. Free.
http://p.sf.net/sfu/Boundary-dev2dev
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to