Fst gives a NoneType error when no indices are given (the default)

[gtsambos]

I have a simulation that does the right thing when I explicitly specify pairs of indices:

ts.Fst(sample_sets = samples_list,
      indexes = [(0,2), (0,1)])

outputs

array([[0.12508121, 0.12780193]])

but that gives the following error if I remove the indexes input (and so use the default value of None)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-6-fbfb01290311> in <module>
----> 1 ts.Fst(sample_sets = samples_list)

/usr/local/lib/python3.7/site-packages/tskit/trees.py in Fst(self, sample_sets, indexes, windows, mode, span_normalise)
   3332         fst = np.repeat(1.0, np.product(divergences.shape))
   3333         fst.shape = divergences.shape
-> 3334         for i, (u, v) in enumerate(indexes):
   3335             denom = (diversities[:, :, u] + diversities[:, :, v]
   3336                      + 2 * divergences[:, :, i])

TypeError: 'NoneType' object is not iterable

[jeromekelleher]

See also #199 (comment)

[jeromekelleher]

I was looking into this @petrelharp, and it seems we don't have a lot of test coverage on Fst right now --- we're only testing the "site" version, essentially. Since this is a derived statistic it seems overboard to implement the full thing from scratch like the other stats, but we do need to get some sort of coverage here on "branch" and "node". Have you any suggestions for this?

[petrelharp]

I guess we just need to do some "does this run" sort of checks. I'll have a look.

[petrelharp]

Currently our default for indexes=None is indexes=[(0, 1)] for pairwise stats. We can't do this here because users might think that the single Fst value returned is the between-all-pops Fst value. I also don't want to just return the full matrix by default because it's not clear what order it should be in. So, I'm making the user specify.

[jeromekelleher]

Currently our default for indexes=None is indexes=[(0, 1)] for pairwise stats. We can't do this here because users might think that the single Fst value returned is the between-all-pops Fst value. I also don't want to just return the full matrix by default because it's not clear what order it should be in. So, I'm making the user specify.

I see. What if [(0, 1)] was the default when we have two sample sets, but we raise a ValueError if there's more than 2 sample sets and indexes is None? We want to make the common case easy and simple.

[petrelharp]

Oh, that's a good idea.

[petrelharp]

I've done this, although similar defaults could be set for many other statistics, which we should probably do at some point...

[jeromekelleher]

I've had a change of heart here, and think maybe we should hold off on specifying a default for this---see the comment in #313

[petrelharp]

Whoops; missed this. Let's discuss at #315 .