[pymvpa] null classification performance in the presence of strong univariate signal??
d.soto.b at gmail.com
Thu Sep 11 15:15:04 UTC 2014
Hi Nick, sorry to come back to this point
but I am not clear why you mention that I have 2 samples
only for MVPA, I have condition A with 19 beta images (1 per subject)
and condition B with the same, merged and concatenated in one fule
If I run a basic nfold crosvalidation
clf = LinearNuSVMC()
sl=sphere_searchlight(cv, radius=3, space='voxel_indices',
then I am training SVM to distinguish condition A from B
in 18 subjects and then testing on the remaining 19th subject.....am i
the univariate stats are done as a A-B t-test on those images....
so why N=2? still did not figure out how the univariate t-test gives
strong signal in frontoparietal cortex but MVPA nothing
On Mon, Sep 8, 2014 at 11:50 AM, Nick Oosterhof <nikolaas.oosterhof at unitn.it
> On Sep 8, 2014, at 12:09 PM, David Soto <d.soto.b at gmail.com> wrote:
> > however I should say that I still do not get why using the very same
> input data, the univariate GLM picks the difference between the*cued and
> uncued* conditions but the MVPA seems not
> From what I understood, you were trying to use the beta estimates /from/
> the GLM as input for MVPA.
> The difference is that the GLM uses N samples (with N the number of
> acquired volumes; i.e. as a timeseries) as input, whereas your MVPA
> approach only uses 2 (which is much smaller than N). Two samples is just
> not enough to do MVPA, just as two samples is insufficient for any
> meaningful univariate statistics.
> You can use all N volumes as input for MVPA, for example see:
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Pkg-ExpPsy-PyMVPA