Gael Varoquaux <gael.varoquaux@...> writes:

> 
> On Thu, Aug 09, 2012 at 01:02:21AM +0000, Abhi wrote:
> >      I am using sklearn.svm.LinearSVC for document classification and I get 
> > a
> >  good accuracy[98%] on predict. Is there a way to find the confidence of 
match
> >  (like predict_proba() in SGDClassifier)?
> 
> Not simply using LinearSVC: liblinear does not implement the mechanism
> need to estimate a confidence interval. You could use
> 'SVC(kernel="linear", probability=True)'.
> 
> HTH,
> 
> Gael
> 

Hi Gael, unfortunately the libSVM implementation is prohibitively slow compared
 to liblinear, which is why I prefer LinearSVC. However, not being able to get 
 the confidence of a prediction is a showstopper. If there's no way to determine
 how closely the input matches the predicted category, how do I 'trust' the
 classification results on on unknown input?

Thanks


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to