[pymvpa] IndexError from gnbsearchlight analysis

[Yaroslav Halchenko]

my blunt guess is that it is due to unbalanced # of samples per each
chunk: i.e. in some data chunk having  no samples of some label.

Could you share

print dataset.summary()

?

and ideally, if more help needed (;-)), dataset itself (h5save it)
so we could reproduce/fix it.

Cheers

On Mon, 25 Jul 2011, Zhen Zonglei wrote:

>    Hi,guys

>    I used the following code to do GNB searchlight analysis,

>    # GNB searchlight analysis

>    dataset = alldata[timeselect,spatialslelect ]
>    sl=
>    sphere_gnbsearchlight(GNB(),NFoldPartitioner(),radius=3,postproc=mean_s
>    ample())
>     res  = sl(dataset)
>    sphere_errors = res.samples[0]

>    BUT,  the errors were reported(see bellow). The version I used is
>    pymvpa  0.6.

>    In addition, with the similar code, the general searchlight analysis
>    can run sucessfully.

>     # Genearal searchlight analysis
>     75                 cv =
>    CrossValidation(linearCSVMC(),NFoldPartitioner(),errorfx=lambda p, t:
>    np.mean(p == t),enabl    e_ca=['stats'])
>     76                 sl =
>    sphere_searchlight(cv,radius=3,postproc=mean_sample())
>     77                 res  = sl(dataset)
>     78                 sphere_errors = res.samples[0]





>    So, what happened in GNB searchlight analysis?



>    Zonglei Zhen





>    -----------------IndexError from GNB searchlight
>    analysis------------------------

>    In [2]: from gnbsearchlight import *
>    Warning: divide by zero encountered in log
>    ERROR: An unexpected error occurred while tokenizing input
>    The following traceback may be corrupted or invalid
>    The error message is: ('EOF in multi-line statement', (43, 0))

>    -----------------------------------------------------------------------
>    ----
>    IndexError                                Traceback (most recent call
>    last)

>    //mystudy/code/python/<ipython console> in <module>()
>    //mystudy/code/python/gnbsearchlight.py in <module>()
>         70                 #GNB searchlight
>    sl =
>    sphere_gnbsearchlight(GNB(),NFoldPartitioner(),radius=3,postproc=mean_s
>    ample())
>    ---> 72                 res  = sl(dataset)
>         73                 sphere_errors = res.samples[0]
>         74

>    /usr/local/neurosoft/lib/python2.6/site-packages/mvpa/base/learner.pyc
>    in __call__(self, ds)
>        220                                    "used and auto training is
>    disabled."
>        221                                    % str(self))
>    --> 222         return super(Learner, self).__call__(ds)
>        223
>        224

>    /usr/local/neurosoft/lib/python2.6/site-packages/mvpa/base/node.pyc in
>    __call__(self, ds)
>         74
>         75         self._precall(ds)
>    ---> 76         result = self._call(ds)
>         77         result = self._postcall(ds, result)
>         78

>    /usr/local/neurosoft/lib/python2.6/site-packages/mvpa/measures/searchli
>    ght.pyc in _call(self, dataset)
>        108
>        109         # pass to subclass

>    --> 110         results, roi_sizes = self._sl_call(dataset, roi_ids,
>    nproc)
>        111
>        112         if not roi_sizes is None:

>    /usr/local/neurosoft/lib/python2.6/site-packages/mvpa/measures/gnbsearc
>    hlight.pyc in _sl_call(self, dataset, roi_ids, nproc)
>        459             norm_weight = -0.5 * np.log(2*np.pi*variances)
>        460             # last added dimension would be for ROIs

>    --> 461             logpriors = np.log(priors[:, np.newaxis,
>    np.newaxis])
>        462
>        463             if __debug__:

>    IndexError: 0-d arrays can only use a single () or a list of newaxes
>    (and a single ...) as an index

> _______________________________________________
> 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


-- 
=------------------------------------------------------------------=
Keep in touch                                     www.onerussian.com
Yaroslav Halchenko                 www.ohloh.net/accounts/yarikoptic