Regularization, Feature Importance for back-to-back regression

[ssaket]

I think it would be useful to have an option of adding regularization to the current back-to-back regression solver. Additionally, if following could be implemented -

• B2B feature importance
• Forward feature importance

[behinger]

have a look at #42 for regularization/CV - I can make b2b much faster, but I haven't found the time yet...

[ssaket]

@behinger I see some new package (dependencies) required for #42 Do we want them? I was thinking of a native solution from scratch.

[behinger]

I'm mostly trying out things. I dislike writing native solutions from scratch because I typically don't have the time to test them through & typically I can't optimize them as much.

But clearly, there are too many dependencies, i.e. one library to do the job should be enough ;-) For me B2B is still very much in research-domain, e.g. I applied it to P300 RT data but it "didnt work", i.e. didnt take RT into account even though univariatly it does.

#42 is still in progress

[behinger]

btw. very welcome to modify my PR

I think you asked for the P3 data somewhere, sorry I forgot.

This is a modified version of your DataAnalysis.jl notebook from the FaPra
https://gist.github.com/f96075ad4ce8632681c87357bd1d3c1a

The p3 data you can find here (careful 700mb): https://cloud.wirdreibei.de/s/i3r2xT9Ep9SC9P9

I didn't double check it, running a bit out of time to do everything I want. If you have trouble just mention it here