[pymvpa] design matrix identical across sessions

[Yaroslav Halchenko]

On Mon, 09 May 2016, Wolfgang Pauli wrote:

> Hi,

> I am trying to perform an mvpa analysis of an experiment in which I have 16
> different trial types and 4 sessions, 4 trial types in each session (run).
> Based on the tutorial, I was getting started by using fit_event_hrf_model,
> like so:

> evds = fit_event_hrf_model(ds, events, time_attr='time_coords',
> condition_attr=('onset'), return_model=True)

> where ds is an openfmri dataset (get_model_bold_dataset).

> I was trying to figure out why I would always get the warning that the
> design matrix was singular, and eventually ended up investigating the
> design matrix of the model.

> I used matplotlib to plot the design matrix, split it up into the four
> session, and found that the four parts were IDENTICAL. What could I be
> doing wrong? Obviously, it shouldn't be the same, because there are
> different trial types in the four sessions, and the trial order is also
> randomized.

> Furthermore, the design matrix has 26 regressors. I don't quite understand
> where that number is coming from, as I have 16 unique event types, and 4
> sessions.


most probably that events definition was was the same in each of the
chunks... but also note that if you do it in a dataset which has
multiple chunks, you want to have condition_attr=['onset', 'chunks'] if
you want to make a model per each onset x chunk pair as was demoed e.g.
at  http://www.pymvpa.org/tutorial_eventrelated.html#response-modeling

if you share your data, and ideally your investigation script then we
could may be look deeper
-- 
Yaroslav O. Halchenko
Center for Open Neuroscience     http://centerforopenneuroscience.org
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834                       Fax: +1 (603) 646-1419
WWW:   http://www.linkedin.com/in/yarik