[pymvpa] Does it make sense to compare SVM weights between two different SVM classifiers?
jetzel at artsci.wustl.edu
Wed Nov 14 20:19:07 UTC 2012
I would be very hesitant to put too much emphasis on interpreting the
svm weight maps, particularly for the purpose of this type of
comparison. One of the reasons is that weight maps in general are
difficult to properly construct and interpret (Lee et al describes some
of the issues).
Besides, svm weight maps are more aimed at the problem of localization
(which voxels are most important for the classification), not comparing
significance of different classifications. Some sort of permutation test
should be able to get at the question of A vs. B > C vs. D.
Rereading, I think you want to describe the classification difference in
terms of characterizing what part of the stimuli lead to the difference
(not which voxels). We've been struggling with this issue: are the
classes *further apart* (centroids more distant, templates more
distinct) in the more-accurate situation, or is the *variance less*
(centroids same but point-clouds tighter)?
For this, we've tried a few different things: PCA (more variance
explained in one case than the other?), looking at the actual voxel-wise
variance levels, measuring distance concentration (see Ata Kaban's
work), visualizing the data. We haven't found a single simple solution,
but it looks like a combination of methods might get us to a convincing
Lee et al. Effective functional mapping of fMRI data with support-vector
machines. DOI: 10.1002/hbm.20955
On 11/14/2012 11:47 AM, Meng Liang wrote:
> Dear MVPA experts,
> In my study, I used the fMRI signals from a given ROI to predict the
> stimulus type for two different classification tasks: (1) type A vs.
> type B, and (2) type C vs. type D (the two classification tasks were
> performed on the same ROI but during different trials: the fMRI data
> used for task 'A vs. B' were taken from trials A and trials B, and the
> data used for task 'C vs. D' were taken from trials C and trials D). It
> was expected that this ROI should provide a higher classification
> accuracy in the task of 'A vs. B' than in the task of 'C vs. D'. The
> results indeed confirmed this. I just wonder whether the higher
> classification accuracy in the task of 'A vs. B' (presumably the higher
> capability of the classifier in task 'A vs. B') relative to the task 'C
> vs. D' could be reflected in the sensitivity maps (i.e., SVM weights) in
> some way? For example, would the SVM of task 'A vs. B' have higher SVM
> weights or a larger margin compared to the SVM of task 'C vs. D'? In
> other words, can I directly compare the sensitivity maps obtained from
> the two different classification tasks?
> I'm not sure if I asked my question clearly. Please let me know if there
> is anything unclear.
> Many thanks!
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
Joset A. Etzel, Ph.D.
Cognitive Control & Psychopathology Lab
Washington University in St. Louis
More information about the Pkg-ExpPsy-PyMVPA