[Eeglablist] SIFT - MVAAR

[Octavian Lie]

Dear All,

I am new to SIFT. I have a 10 sec baseline multivariate time series (m
channels, t time samples), and 3 seizures (3x 10 sec seizure epochs
(annotated by an epileptologist) in the same patient). I study
causality/propagation, and not prediction.
1. I wanted to compare the baseline against each of the 3 seizures, and
seizures among themselves (pairwise, not all together), to see to what
degree baseline differs from/ is similar to seizures and seizures among
themselves. What would be a good test offered by SIFT for this
(esentially repeated-measures two multivariate (m) sample test
with an unknown distribution fo the m variable values); I was told LOOM
(leave one out) crossvalidation is one option, but I do seem to see that
offered.
2. From the MVAAR models in SIFT, is there one that has shown to work well
(better than others) on high-dimensional and nonstationary data of the kind
seen with ictal ecog? Since I do not study prediction, overfitting the
model by increasing m should not matter, is it right? If not, is there an
overfitting measure offered?

Please advise,

Viorel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20141208/721d0063/attachment.html>