[pymvpa] Confusion matrix and boxcars

[Per B. Sederberg]

On Sun, Sep 14, 2008 at 10:23 AM, Yaroslav Halchenko
<debian at onerussian.com> wrote:
> I thought about ROC... but haven't thought it through yet -- problem
> comes with multiclass classification -- there are ways to pull in ROCs
> in there but needs thinking.

Yes, I've read a couple papers on multi-class AUC calculations.  It
turns out it's not too hard.

>
> event related -- imho we just need ability to append 1 feature into the
> 'mapping' - offset (in sec) from first volume onset which got into
> mapped 'sample'. That would account for what we need I think

Great idea!  I have a ton of data with which to try it out (i.e., all
my data).  I look forward to chattin' more about it soon.

P

>
> On Sun, 14 Sep 2008, Per B. Sederberg wrote:
>
>> Howdy All:
>
>> How hard would it be to add the area under the ROC (AUC) to the
>> confusion matrix?  The method is already in pymvpa, but one issue is
>> that you need to pass the predicted values (not the actual
>> predictions.)  We've talked about this a bit in the past.
>
>> Also, I'd like to explore how we should deal with event-related fMRI
>> data where the timing of the events do not line up with the start of a
>> TR.  The BoxcarMapper seems to be the way to go, but it'll take some
>> thinking to deal with random event offsets.
>
>> Perhaps we can chat about this next week after the paper is out.
>
>> Latros,
>> P
>
>> _______________________________________________
>> Pkg-ExpPsy-PyMVPA mailing list
>> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
>> http://lists.alioth.debian.org/mailman/listinfo/pkg-exppsy-pymvpa
>
>
> --
> Yaroslav Halchenko
> Research Assistant, Psychology Department, Rutgers-Newark
> Student  Ph.D. @ CS Dept. NJIT
> Office: (973) 353-5440x263 | FWD: 82823 | Fax: (973) 353-1171
>        101 Warren Str, Smith Hall, Rm 4-105, Newark NJ 07102
> WWW:     http://www.linkedin.com/in/yarik
>