[pymvpa] Z-scoring raw data

[Derek Huffman]

Hello all,

I noticed that in the tutorial data that the data is z-scored with respect
to the resting condition. I have not seen this done in the literature and I
am I am wondering if you have found it beneficial to z-score in that way or
if just doing a simple z-score on the entire time course (per run) provides
similar classification accuracy.

Using the tutorial data and the vt mask with LinearCSVMC classifier, I
z-scored by just doing zscore(ds) (instead of zscore(ds,
param_est=['targets',['rest']))) and it appeared to actually have better
accuracy than z-scoring on the rest data. I am currently designing an
experiment with a scan session that is over an hour and a half and I am
wondering if I need to collect a bunch of resting trials for z-scoring
(which will make my scan even longer) or if I would be safe to assume that
z-scoring the time series will be sufficient. My design has 4 second trial
presentations and 8 second ITI's (12 second stimulus onset asynchrony).

Your help is greatly appreciated! Thank you!
Derek Huffman
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20120419/e2aefa15/attachment.html>