hi alex,

what you observe seem to be a huge variance of the estimator.
>


> I would smooth the fMRI data (I assume it is) to stabilize the anova
> feature
> selection.
>

unfortunately these are measures associated with structural features of the
brain that are matched across subjects. i could try normalizing by brain
volume or something.

subject (28) x features (300)

cheers,

satra


>
> let me know if this helps
>
> Alex
>
> On Wed, Mar 14, 2012 at 9:27 PM, Satrajit Ghosh <[email protected]> wrote:
> > i have a situation i cannot quite comprehend. any suggestions would be
> much
> > appreciated.
> >
> > ---
> > clf = sk.svm.SVC(kernel='linear', shrinking=True)
> > anova_filter = SelectKBest(f_regression, k=50)
> > clf = Pipeline([('anova', anova_filter), ('svc', clf)])
> > maxkfold = np.min(np.bincount(y1)[1:])
> > result = []
> > for train, test in cv.StratifiedKFold(y1, maxkfold):
> >     result.append((y1[test],
> >                   clf.fit(X1[train], y1[train]).predict(X1[test])))
> > ---
> >
> > with k=50 in the above anova filter i get the following confusion matrix
> >
> > [[ 0  7]
> >  [ 0 10]]
> >
> >
> > with k=51 and greater, i get ( i love this - but i don't have much
> > confidence in it):
> >
> > [[7 0]
> >  [1 9]]
> >
> > if i throw a Normalizer in the pipeline:
> >
> > clf = Pipeline([('xfm', Normalizer()), ('anova', anova_filter), ('svc',
> > clf)])
> >
> > then i get the pattern with k=50 above independent of what i set k to.
> >
> > if i do shufflesplit it results in the same issues.
> >
> > cheers,
> >
> > satra
> >
> >
> >
> ------------------------------------------------------------------------------
> > Virtualization & Cloud Management Using Capacity Planning
> > Cloud computing makes use of virtualization - but cloud computing
> > also focuses on allowing computing to be delivered as a service.
> > http://www.accelacomm.com/jaw/sfnl/114/51521223/
> > _______________________________________________
> > Scikit-learn-general mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
> >
>
>
> ------------------------------------------------------------------------------
> Virtualization & Cloud Management Using Capacity Planning
> Cloud computing makes use of virtualization - but cloud computing
> also focuses on allowing computing to be delivered as a service.
> http://www.accelacomm.com/jaw/sfnl/114/51521223/
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
------------------------------------------------------------------------------
Virtualization & Cloud Management Using Capacity Planning
Cloud computing makes use of virtualization - but cloud computing 
also focuses on allowing computing to be delivered as a service.
http://www.accelacomm.com/jaw/sfnl/114/51521223/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to