[pymvpa] classifier prediction question

[Yaroslav Halchenko]

look into a classifier ca.estimates.  Some classifiers (e.g. SMLR, GNB)
would base their decision on e.g. a posterior probability which would
then be stored in the clf.ca.estimates for a classifier clf upon making
a prediction. E.g. 

In [18]: clf = mv.SMLR(enable_ca=['predictions'])

In [19]: clf.train(mvtd.datasets['uni3small'])

In [20]: clf.predict(mvtd.datasets['uni3small'])
Out[20]: 
array(['L0', 'L0', 'L0', 'L0', 'L0', 'L0', 'L0', 'L0', 'L0', 'L0', 'L0',
       'L0', 'L1', 'L1', 'L1', 'L1', 'L1', 'L1', 'L1', 'L1', 'L1', 'L1',
       'L1', 'L1', 'L2', 'L2', 'L2', 'L2', 'L2', 'L2', 'L2', 'L2', 'L2',
       'L2', 'L2', 'L2'], 
      dtype='|S2')

In [21]: print clf.ca.estimates
[[  9.98840082e-01   7.72142962e-04   3.87774658e-04]
 [  9.97071204e-01   2.78187822e-03   1.46917290e-04]
 [  9.89887463e-01   4.86107005e-03   5.25146706e-03]
 [  9.96544159e-01   1.23337390e-03   2.22246665e-03]
 [  9.76508361e-01   2.31793063e-03   2.11737084e-02]
 [  8.52440274e-01   4.06182039e-02   1.06941522e-01]
 [  9.99943827e-01   1.12451619e-05   4.49279579e-05]



On Thu, 16 Jan 2014, Jason Ozubko wrote:

>    Perhaps a very newbie question but when you call clf.predict is it
>    possible to have the function return more than just a single prediction?
>    As in, if I have 4 target labels, is it possible to get, for each test
>    sample, the probability (or some other metric) with which the classifier
>    thinks that each of those 4 target labels apply?
>    So for example, if you had target types of "animal", "vegetable",
>    "mineral", and "person" and you trained up a classifier, then with
>    clf.predict I could submit a handful of test samples and get results like�
>    ["vegetable"
>    "vegetable"
>    "animal"
>    "person"
>    "mineral"
>    "mineral"]
>    But is there any way to instead get a read out that says something like,
>    for the first sample the classifier would have picked vegetable first,
>    then animal, then person, and lastly mineral. �For the second sample
>    however the classifier would have picked vegetable then person, then
>    animal, then mineral? �So I could see not only what option the model
>    predicts but also how close was each test sample to the other options as
>    well?
>    Thanks in advance
>    Cheers,
>    Jason

> _______________________________________________
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa


-- 
Yaroslav O. Halchenko, Ph.D.
http://neuro.debian.net http://www.pymvpa.org http://www.fail2ban.org
Senior Research Associate,     Psychological and Brain Sciences Dept.
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