On Sun, Mar 18, 2012 at 4:45 AM, James Bergstra
<[email protected]> wrote:

> I agree that it's a good idea to correct C for sample size when moving
> from a sub-problem to the full thing.  I just wouldn't use the word
> "optimal" to describe the new value of C that you get this way - it's
> an extrapolation, a good guess... possibly provably better than the
> un-corrected value of C, but I would balk at claiming that it's
> optimal.

Completely agreed. The assumption that the relation between the
training set size and the optimal C value is linear may not hold in
some cases (for example, if the data lies in highly dense regions and
sparse regions).

I also agree that the only safe way to tune C is to use held-out data:
this way we don't need to refit the best model.

> The alpha specified this way could (should?) have the same name and
> interpretation as the l2_regularization coefficient in the
> SGDClassifier.

Would you convert alpha into a C internal value or would you patch
libsvm / liblinear to use alpha? I don't understand how the former
would be different from the scale_C option, in practice.

Mathieu

------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here 
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to