Hi,
>Thanks! Could you also run a quick grid search to fine tune gamma and
>C for all models independently?
I have updated the code on the gist to match your requirement. I made a 5
fold splits where on each the parameters are finely estimated for both KPLS
and SVC using gridsearchCV.
In conclusion the results of KPLS works similar (3 times) or better than
SVC (2 times out of 5). Please feel free to comment on the setup or the
experiments I followed.
>Another question: what is the time complexity for KPLS w.r.t.
>n_samples? quadatratic, cubic?
Currently it should be cubic as I am using SVD to estimate the eigen
vectors of the kernel matrix. However, I think this can be optimized by
taking other approaches.
>Could you please extend your gist to randomly subsample the digits
>dataset to only keep 50% and re-run CV for the 3 models on that?
Already made that, its all on the public gist.
Regards,
Eweiwi
On Thu, Dec 5, 2013 at 2:20 PM, Olivier Grisel <olivier.gri...@ensta.org>wrote:
> 2013/12/5 abdalrahman eweiwi <abdalrahman.ewe...@gmail.com>:
> > Hi,
> >
> > I have a problem of getting GridSearchCV working with kpls for fine
> tunning
> > its parameters. I have reviewed the documentation for how to get a new
> > estimator work with GridSearchCV but I did not find enough information
> > there. I pursued myself on this but finally I got stuck with a problem
> > related to multi-class labels and the cross validation. As
> cross-validation
> > randomly subsample from all the data, it may take distinct number of
> classes
> > at each iteration, but I need the sampler to get random samples for each
> > class so that all classes are represented at each iteration. Is there a
> way
> > to go around this problem?
>
> You should not have the problem if you use StratifiedKFold (the
> default for classification) or StratifiedShuffleSplit.
>
>
> --
> Olivier
> http://twitter.com/ogrisel - http://github.com/ogrisel
>
>
> ------------------------------------------------------------------------------
> Sponsored by Intel(R) XDK
> Develop, test and display web and hybrid apps with a single code base.
> Download it for free now!
>
> http://pubads.g.doubleclick.net/gampad/clk?id=111408631&iu=/4140/ostg.clktrk
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-general@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
------------------------------------------------------------------------------
Sponsored by Intel(R) XDK
Develop, test and display web and hybrid apps with a single code base.
Download it for free now!
http://pubads.g.doubleclick.net/gampad/clk?id=111408631&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
