Ronan Perry

@agramfort I have spoken with @idc9 and he is starting to make PRs with these methods/updates.

Yes I think this could be useful. I've run into such an issue before applying groupPCA like algorithms to many correlation matrices at once. I'm not very familiar with memory...

Very strange, it doesn't give me that result locally. There shouldn't be any randomness in the script, and it gives me consistent results locally.

I believe that in order to make this work, github.com/mvlearn/mvlearn.github.io needs to have folders with the saved docs for each version we wish the website to render (see [mne-tools](https://github.com/mne-tools/mne-tools.github.io) with...

Our webdocs are rendered differently though as part of the organization, not just a single repository. After a PR, circleci compiles the web documentation from `mvlearn` and pushes it to...

Is it possible to just use [itertools.product](https://docs.python.org/3/library/itertools.html#itertools.product) and pass in the mvlearn CV splitter to the sklearn Gridsearch?

Yeah I struggled to find settings for sphinx but I'd love to be proven otherwise. I know that we use a non-default scheme so that may not help.

Yeah I use sphinx_rtd_theme but wasn't able to find settings options. The closest was an installable [https://github.com/AaltoSciComp/sphinx_rtd_theme_ext_color_contrast](https://github.com/AaltoSciComp/sphinx_rtd_theme_ext_color_contrast) which makes some changes to the base sphinx which is a valid option.

Or rather than matplotlib, you could draw a pointcloud `G` (with some edges based on the G shape) and graph match an actual graph to it, using the matched pointcloud...

In the paper they use a decision tree to compute the mean squared error, but I don't think there is anything stopping any other method. Thus the implementation may want...