[pymvpa] Dataset with multidimensional feature vector per voxel

Ulrike Kuhl kuhl at cbs.mpg.de
Tue Nov 24 17:57:37 UTC 2015 

Dear all,

thanks so much for your help so far! Unfortunately I've run into another issue in my analysis (sorry to be such a nuisance...):

When I started to run my script on real data I got an 'index out of bounds' exception during searchlighting. 
I've figured out that it has to do with the usage of a mask - in my toy runs, I did not include a mask and everything was fine. When looking at real data I definitely want to use a mask to restrict analysis only to within brain voxels or specific ROIs. 
It seems that the searchlight searched at voxels not included in the mask - hence yielding an 'IndexError' when it runs out of the mask. When I use a whole brain mask with a volume of 1399939 voxels I get:

Traceback (most recent call last):
  File "AGLmvpa_small.py", line 154, in <module>
    res = sl(DS)
  File "/usr/lib/python2.7/dist-packages/mvpa2/base/learner.py", line 259, in __call__
    return super(Learner, self).__call__(ds)
  File "/usr/lib/python2.7/dist-packages/mvpa2/base/node.py", line 121, in __call__
    result = self._call(ds)
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 143, in _call
    results = self._sl_call(dataset, roi_ids, nproc)
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 371, in _sl_call
    results=self.__handle_all_results(p_results))
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 207, in _concat_results
    results = sum(results, [])
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 527, in __handle_all_results
    for r in results:
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 764, in next
    self.store()
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 400, in store
    self.store_data(channel)
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 747, in store_data
    data = channel.receive()
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 135, in receive
    obj = self._receive()
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 121, in _receive
    raise obj
IndexError: index 2231037 out of bounds 0<=index<1399939

I was able to reproduce the error when using a mask (1736 voxels) on my toy data:

Traceback (most recent call last):
  File "DummyMvpa_noisy.py", line 322, in <module>
    res_noisy = sl(DS_noisy)
  File "/usr/lib/python2.7/dist-packages/mvpa2/base/learner.py", line 259, in __call__
    return super(Learner, self).__call__(ds)
  File "/usr/lib/python2.7/dist-packages/mvpa2/base/node.py", line 121, in __call__
    result = self._call(ds)
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 143, in _call
    results = self._sl_call(dataset, roi_ids, nproc)
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 371, in _sl_call
    results=self.__handle_all_results(p_results))
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 207, in _concat_results
    results = sum(results, [])
  File "/usr/lib/python2.7/dist-packages/mvpa2/measures/searchlight.py", line 527, in __handle_all_results
    for r in results:
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 764, in next
    self.store()
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 400, in store
    self.store_data(channel)
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 747, in store_data
    data = channel.receive()
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 135, in receive
    obj = self._receive()
  File "/usr/lib/pymodules/python2.7/pprocess.py", line 121, in _receive
    raise obj
IndexError: index 1736 out of bounds 0<=index<1736


Why does the searchlight run out of the mask?

Thanks again for your help!
Ulrike


----- Original Message -----
From: "Bill Broderick" <billbrod at gmail.com>
To: "pkg-exppsy-pymvpa" <pkg-exppsy-pymvpa at lists.alioth.debian.org>
Sent: Thursday, 19 November, 2015 16:15:05
Subject: Re: [pymvpa] Dataset with multidimensional feature vector per voxel

No problem! I've only used the random majority under-sampling with replacement so far, but it has a whole lot of options. 

On Thu, Nov 19, 2015 at 10:01 AM, Yaroslav Halchenko < debian at onerussian.com > wrote: 



On Thu, 19 Nov 2015, Bill Broderick wrote: 

> I ran into a similar issue with unbalanced classification and wanted 
> to look at the individual partitions as well. I couldn't figure 
> out how to do so just in PyMVPA, so I ended up using a separate Python module, 
> UnbalanceDataset: https://github.com/fmfn/UnbalancedDataset . With 
> that, I sub-sampled the more common group to balance the two groups, 
> which created a new dataset. I was then able to investigate what was 
> going on in that dataset and what each of the partitions look like as 
> if it were a regular dataset. 

That is a sweet little toolbox -- thanks for sharing! 

-- 
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 

_______________________________________________ 
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 


_______________________________________________
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
-- 
Max Planck Institute for Human Cognitive and Brain Sciences 
Department of Neuropsychology (A219) 
Stephanstraße 1a 
04103 Leipzig 

Phone: +49 (0) 341 9940 2625 
Mail: kuhl at cbs.mpg.de 
Internet: http://www.cbs.mpg.de/staff/kuhl-12160

More information about the Pkg-ExpPsy-PyMVPA mailing list