[Pkg-exppsy-maintainers] Bug in NiftiDataset possibly fixed

[Michael Hanke]

Hi folks,

On Wed, Feb 13, 2008 at 01:59:44AM -0500, Yaroslav Halchenko wrote:
> Hi Per!
> 
> Thanks a lot of looking in depth at the situation ;-)
Seconded. Thanks a lot for taking care. It drove me mad yesterday...

> Explicit default values were set for the convenience of the user who
> could simply '?' method in ipython to get instant idea of what default
> parameters are.
Right, but given the facts Per pointed to, I think the price for this
convenience is too high. It adds a level of obscurity that I cannot
comprehend ;-)

> I haven't yet looked into the changes (last 2 days were heck of
> experience... a kid got sick... ER etc...) -- will do tomorrow, but I
> wanted to share my idea about other default parameters handling we have
> now -- primarily in obscured classifiers like Proxy and combined
> classifier.
> 
> I think we should come up with a generic collection of default arguments
> (plain dict can serve as a starting point). This way it will be
> centralized, thus any user who wants to tune defaults through-out the
> code can do that easily. I will come up with more details and more
> reasoning about that later (after I enter back in the pymvpa development
> world ;))
Cool!

For now, I removed all dangerous default arguments (see my recent
commits). So wherever we go from here, we start on safe ground.

I also finally killed the 'callables=[]' thingie. I wanted to do that
for all while now, but this was the perfect opportunity (reasoning is in
the commit messages).


It looks like we are already in the middle of a bug triage ;-)

BTW: I added two more examples. One of them is a light version of the
searchlight example we already had. The former one is more of an example
about how to code command line tools with PyMVPA (and it is currently
broken BTW).

One week to go.


Best,

Michael



-- 
GPG key:  1024D/3144BE0F Michael Hanke
http://apsy.gse.uni-magdeburg.de/hanke
ICQ: 48230050