[pymvpa] RSA/RDM cross-validated mahalanobis

[basile pinsard]

In fact the method implemented in PyMVPA is not similar to what is used in
the paper.
The CDist Node will compute distances between all pairs of targets betas b
(eg. i,j) between the 2 chunks (A/B), that is for squared euclidian:
d(b_iA,b_jB) = (b_iA-b_jB).(b_iA-b_jB)T
While the crossvalidated squared euclidian as described in this paper is:
d_AB(b_i,b_j) = (b_iA-b_jA).(b_iB-b_jB)T .
The first measures distance between patterns across chunks.
The second measures between chunks "covariance" of intra-chunk multivariate
distance which are then tested for ranks or ordinal stability across
cross-validation chunks.

I will try to implement the latter in PyMVPA at some point.

On Tue, Apr 25, 2017 at 4:23 PM, Matteo Visconti di Oleggio Castello <
matteo.visconti at gmail.com> wrote:

> Yes, you should be able to use it within searchlights as any other Measure
> class. Unfortunately, I do not think that it can support a varying
> covariance matrix at the moment, which makes it less ideal to work with
> Mahalanobis distances. We should augment that to allow such option.
>
> On Tue, Apr 25, 2017 at 12:44 PM, basile pinsard <basile.pinsard at gmail.com
> > wrote:
>
>> Hi Matteo,
>>
>> thanks, that's great, I will give it a try.
>> Also do you think this can be used in a Searchlight?
>> Even with varying kwargs to give the spatial covariance matrix in case of
>> mahalanobis?
>>
>> Thank you.
>>
>> Basile
>>
>> On Tue, Apr 25, 2017 at 12:13 PM, Matteo Visconti di Oleggio Castello <
>> matteo.visconti at gmail.com> wrote:
>>
>>> Hi Basile,
>>>
>>> the new release of PyMVPA 2.6.1 introduces mvpa2.measures.rsa.CDist,
>>> that allows you to compute cross-validated RSA (
>>> https://github.com/PyMVPA/PyMVPA/blob/master/mvpa2/measures/rsa.py#L27),
>>> and you can specify which distance metric you want to use, with additional
>>> kwargs.
>>>
>>> Matteo
>>>
>>> On Tue, Apr 25, 2017 at 11:14 AM, basile pinsard <
>>> basile.pinsard at gmail.com> wrote:
>>>
>>>> Hi PyMVPA developers and users.
>>>>
>>>> I was wondering if there was a way in PyMVPA to use the approaches
>>>> described in this paper:
>>>> "Reliability of dissimilarity measures for multi-voxel pattern analysis"
>>>> were the use of cross-validated mahalanobis distance is advocated in
>>>> favor of correlation based RSA measures.
>>>>
>>>> Thank you.
>>>>
>>>> --
>>>> Basile Pinsard
>>>>
>>>> *PhD candidate, *
>>>> Laboratoire d'Imagerie Biomédicale, UMR S 1146 / UMR 7371, Sorbonne
>>>> Universités, UPMC, INSERM, CNRS
>>>> *Brain-Cognition-Behaviour Doctoral School **, *ED3C*, *UPMC, Sorbonne
>>>> Universités
>>>> Biomedical Sciences Doctoral School, Faculty of Medicine, Université de
>>>> Montréal
>>>> CRIUGM, Université de Montréal
>>>>
>>>> _______________________________________________
>>>> 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
>>>>
>>>
>>>
>>>
>>> --
>>> Matteo Visconti di Oleggio Castello
>>> Ph.D. Candidate in Cognitive Neuroscience
>>> Dartmouth College
>>>
>>> +1 (603) 646-8665 <(603)%20646-8665>
>>> mvdoc.me <http://www.mvdoc.me/> || linkedin.com/in/matteovisconti ||
>>> github.com/mvdoc
>>>
>>> _______________________________________________
>>> 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
>>>
>>
>>
>>
>> --
>> Basile Pinsard
>>
>> *PhD candidate, *
>> Laboratoire d'Imagerie Biomédicale, UMR S 1146 / UMR 7371, Sorbonne
>> Universités, UPMC, INSERM, CNRS
>> *Brain-Cognition-Behaviour Doctoral School **, *ED3C*, *UPMC, Sorbonne
>> Universités
>> Biomedical Sciences Doctoral School, Faculty of Medicine, Université de
>> Montréal
>> CRIUGM, Université de Montréal
>>
>> _______________________________________________
>> 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
>>
>
>
>
> --
> Matteo Visconti di Oleggio Castello
> Ph.D. Candidate in Cognitive Neuroscience
> Dartmouth College
>
> +1 (603) 646-8665 <(603)%20646-8665>
> mvdoc.me <http://www.mvdoc.me/> || linkedin.com/in/matteovisconti ||
> github.com/mvdoc
>
> _______________________________________________
> 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
>



-- 
Basile Pinsard

*PhD candidate, *
Laboratoire d'Imagerie Biomédicale, UMR S 1146 / UMR 7371, Sorbonne
Universités, UPMC, INSERM, CNRS
*Brain-Cognition-Behaviour Doctoral School **, *ED3C*, *UPMC, Sorbonne
Universités
Biomedical Sciences Doctoral School, Faculty of Medicine, Université de
Montréal
CRIUGM, Université de Montréal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20170427/13241ad1/attachment-0001.html>