hyperalignment isn't working correctly

[jeremymanning]

As we explored today, hyperalignment in both the Python and Matlab toolboxes seems to be broken. I propose replacing our implementations with existing implementations that are already known to work. One option is to use the SRM model in the BrainIAK toolbox: https://github.com/IntelPNI/brainiak/blob/master/brainiak/funcalign/srm.py

Another option is to use the Hyperalignment implementation from the PyMVPA2 toolbox: https://github.com/PyMVPA/PyMVPA/blob/master/mvpa2/algorithms/hyperalignment.py

It seems like the SRM/BrainIAK option would be easier, since the toolbox is smaller and less complicated (and the necessary function seems to rely on just a single file, with dependencies that we already have for other parts of our toolbox).

Thoughts?

[jeremymanning]

Note: the matlab version is also broken (in the same way as they python version), and it should be scrapped/replaced.

[andrewheusser]

I agree that we should just port over a python implementation. The intel version would be nice since its compact, but could potentially be more work to get going? Another option would be to ask the Haxby lab if they have a stand alone python implementation, or just rearrange the one they have to remove dependencies on other parts of pymvpa

[jeremymanning]

hey @swaroopgj -- i've added you to this issue to help track it. thanks very much for your help!

[swaroopgj]

@andrewheusser @KirstensGitHub do you happen to have some data or code with which I can re-create your bug?

[andrewheusser]

@swaroopgj yes, try running the examples/hypertools_demo-align_test.py

thanks for your help!

[jeremymanning]

from scipy.io import loadmat as load from mvpa2.suite import * import hypertools as hyp

data = load('test_data.mat') data = [data['spiral'], data['randwalk']]

aligned1 = hyp.align(data)

ha = Hyperalignment() mappers = ha(map(lambda x: Dataset(x), data)) aligned2 = [m.forward(d) for m, d in zip(mappers, data)]

test_data.mat.zip

[jeremymanning]

^ we want aligned1 (our implementation) and aligned2 (mvpa2 implementation) to be the same.

(note: download and unzip test_data.mat.zip)

[swaroopgj]

Sent a PR to fix the issue. @jeremymanning @andrewheusser Close this issue if you think it addresses this issue adequately.

[andrewheusser]

LGTM! @jeremymanning - the hyperalignment code appears to be working :) @swaroopgj added an example in hypertools_demo-align_test.py if you want to check it out.

It looks like we can still optimize the alignment (e.g. multiple iterations for step 2, regularizing at various steps) but the basic algorithm seems to be working! Thanks @swaroopgj!!