IncrementalPCA should get closer to "true" PCA as the number of components increases - so if anything the solution should be more stable rather than less. The difference mostly lies in the incremental processing - regular PCA with reduced components performs the full PCA, then only keeps a subset of the components whereas the incremental version has to slice each sub-decomposition, resulting in a slightly different solution. When all components are kept, the results should be identical (at least in all cases I tested).
You might try using a feature selector such as shown in this example http://scikit-learn.org/stable/auto_examples/feature_selection/plot_rfe_with_cross_validation.html with the 1000 component version. I don't know anything about OCSVMin particular but this is generally useful when you add features and things start to perform much worse. On Wed, Oct 14, 2015 at 9:18 AM, Luca Puggini <lucapug...@gmail.com> wrote: > Thanks for the answer. > I was expecting OCSVM to be not too much influenced by the increasing number > of variables even if some of them are irrelevant. > I am just wondering if the drop in performances is more likely to occur due > to the overfitting of OCSVM or due to an unexpected behaviour of of > Incremental PCA when the number of components is large. > > For the moment the first one seems to be more likely. > > Please write me if you have other opinions. > > Thanks, > Luca > > > On Wed, Oct 14, 2015 at 1:02 PM olologin <ololo...@gmail.com> wrote: >> >> On 10/14/2015 02:28 PM, Oliver Tomic wrote: >> >> I am not sure whether there is such a feature in scikit-learn, but the >> cumulative (validated) explained variance after each component may also give >> a good indication of when to stop including further components. that is when >> it starts to drop. >> >> explained_variance_ratio_ attribute? >> >> ------------------------------------------------------------------------------ >> _______________________________________________ >> Scikit-learn-general mailing list >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > -- > > Sent by mobile phone > > > ------------------------------------------------------------------------------ > > _______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > ------------------------------------------------------------------------------ _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
