PC-Relate implementation

[ravwojdyla]

Relevant discussion: https://discourse.smadstatgen.org/t/pc-relate-experiment/
Code: https://github.com/related-sciences/gwas-analysis/tree/rav-pc_relate

Potential plan could be:

1. move the code+tests over here with adjustments for the sgkit API
2. improve upon the implementation (this could be tracked as a separate issue(s))

We can use this issue to track the work necessary to move it over here.

[ravwojdyla]

@tomwhite I know you have started reviewing the PC-Relate code - what's the state right now?

[tomwhite]

My focus was on the locating the performance bottlenecks in the PC-Relate code (brief notes here: https://github.com/related-sciences/gwas-analysis/blob/rav-pc_relate-tw/notebooks/platform/xarray/pc_relate_perf.py#L3-L7).

I think your plan makes sense - move the code over to this repo once the sgkit API becomes a bit clearer.

[ravwojdyla]

Thanks @tomwhite I see. Regarding perf/gramian issue, I thought it might also be interesting to try to represent it with optimised einsum, plus you've have mentioned potentially computing only the parts we are interested in, which is equally interesting! I bet there is lot we can improve - but yea, let's first move it in, and then optimise.

[ravwojdyla]

FYI: I will go ahead and create a PR for the current state.

[ravwojdyla]

Closing this via #228