[pymvpa] Generating more 'realistic' test data?

[Shane Hoversten]

Hi -

The data_generators module (
http://www.pymvpa.org/generated/mvpa2.misc.data_generators.html) has lots
of methods to generate various flavors of test data, but the ones I
actually understand seem to produce a rather simple sort of multivariate
data, where differences between classes are characterized by modifications
to a single diagnostic nonbogus feature for each class, s.t. that single
diagnostic voxel has a larger mean.

I'd like to do something like generate data where condition A lives in some
subspace of the feature space, and condition B lives in another subspace,
where the subspaces can span larger regions than the 2d ones like in
pure_multivariate_signal; or live on a certain simplex; or something else
where classes live within a more sophisticated high-dimensional decision
boundary.

It's possible that such a thing can be assembled with the tools already
there and I just don't see how; if so, can someone suggest how it might be
accomplished?  Or is this a weird-enough thing to want that I need to roll
my own?

Thanks,

Shane
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20140903/36eefe4b/attachment.html>