Tidy ups for genetic_relatedness

[jeromekelleher]

From @petrelharp's comments here in #898:

• I think the denominator should always be 2, and should be incorporated into the summary function.
• I think the C and python methods should be consistently named: one is relatedness, the other genotype_relatedness; I think we decided to do the second thing.

We also want to merge the current tests for genetic_relatedness into the existing framework for stats tests. Hopefully this can be done by following the pattern for something like trait regression

I'd imagine that we'd basically be moving everything out of the current test_covariance.py module, but maybe not.

[edit testing stuff moved to #974]

How does this sound @brieuclehmann?

[brieuclehmann]

Sorry, I messed around with the names while we were switching from a branch to a site view!

We had agreed on genetic_relatedness initially. (I changed to genotype_relatedness when thinking about sites.) Happy to change the names.

Would it be worth opening a separate issue for the tests? There are a few more to be written (e.g. for the 'branch' and 'node' modes).

[jeromekelleher]

Would it be worth opening a separate issue for the tests? There are a few more to be written (e.g. for the 'branch' and 'node' modes).

Yes, good idea. Would you mind doing this?

[petrelharp]

No worries about the switching around! That's what happens in development. =) I still vote for genetic_relatedness, since the name should also describe mode="branch" as well as the genotype statistics.