[pymvpa] hyperalignment inquiry

[Swaroop Guntupalli]

The hyperalignment example on PyMVPA uses one beta map for each category
per run.

On Wed, Jul 27, 2016 at 2:57 PM, Swaroop Guntupalli <swaroopgj at gmail.com>
wrote:

> Hi David,
>
> Beta maps should work fine for hyperalignment. The more maps (or TRs)
> there are, better the estimate.
> We used within-subject hyperalignment in Haxby et al. 2011, which uses
> maps from 6 categories (we used 3 successive betas per condition I think).
>
> vstack() merges multiple datasets into a single dataset, and if there is
> any voxel count (nfeatures) mismatch across subjects, it won't work (as
> evidenced by the error).
> Hyperalignment takes in a list of datasets, one per each subject.
> So, you can make that a list as
> ds_all =[ds1, ds2, ...., ds16]
> and use for Hyperalignment()
>
> Best,
> Swaroop
>
>
> On Wed, Jul 27, 2016 at 2:28 PM, David Soto <d.soto.b at gmail.com> wrote:
>
>> hi,
>>
>> in my experiment I have 28 betas in condition A and 28 parameter estimate
>> images and 28  in condition B for each subject (N=16 in total).
>>
>> i have performed across-subjects SVM-based searchlight classification
>> using MNI-registered individual beta images and I would like to repeat and
>> confirm my results using searchlight based on hyperaligned data.
>>
>> i am not aware of any paper using hyperaligment on  beta images but I
>> think this should be possible, any advise please would be nice
>>
>> i've created individual datasets concatenating the 28 betas in condition
>> A and the 28 in condition (in the actual experiment condition A and B can
>> appear randomly on each trial). I have 16 nifti datasets, one per subject,
>> with each in individual native anatomical space. In trying to get a dataset
>> in the same format as in the hyperlignment tutorial I use fmri_dataset on
>> each individual wholebrain 48 betas  and then try to merged then all i.e. ds_merged
>> = vstack((d1, d2, d3, d4, d5, d6, d7, d8, d9, d10, d11, d12, d13, d14,
>> d15,d16)) but this gives the following error pasted at the end,
>> which I think it is becos the number of voxels is different across
>> subjects. This is one issue.
>>
>> Another is that the function vstack does appear to produce the list of
>> individual datasets that is in the hyperligment tutorial dataset, but a
>> list of individual betas, I would be grateful to receive some tips.
>>
>> thanks!
>> david
>> ------------------------------------------------------------
>> ---------------
>> ValueError                                Traceback (most recent call
>> last)
>> <ipython-input-64-2fef46542bfc> in <module>()
>>      19 h5save('/home/dsoto/dsoto/fmri/wmlearning/h5.hdf5', [d1,d2])
>>      20 #ds_merged = vstack((d1, d2, d3, d4, d5, d6, d7,d8,d9, d10, d11,
>> d12, d13, d14, d15, d16))
>> ---> 21 ds_merged = vstack((d1, d2))
>>
>> /usr/local/lib/python2.7/site-packages/mvpa2/base/dataset.pyc in
>> vstack(datasets, a)
>>     687                              "datasets have varying attributes.")
>>     688     # will puke if not equal number of features
>> --> 689     stacked_samp = np.concatenate([ds.samples for ds in
>> datasets], axis=0)
>>     690
>>     691     stacked_sa = {}
>>
>> ValueError: all the input array dimensions except for the concatenation
>> axis must match exactly
>>
>> _______________________________________________
>> 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
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20160727/0ceb56dc/attachment-0001.html>