Ancestral state polarisation and Peter Keightley's est-sfs software

[hyanwong]

When using tsinfer, I think that a lot of people are going to get stuck on the ancestral state polarisation requirement. @ab08028 pointed out Peter Keightley's GPLed est-sfs software (see http://www.homepages.ed.ac.uk/pkeightl/software.html) and this seems like the best recommendation to me.

At a minimum, I think it would be sensible to note this in the tsinfer documentation. But I wonder if we could make it easier for users somehow, e.g. if in the long run (when we change tsinfer to rely on sgkit storage) there's a way for sgkit to output data in the format required by est-sfs, and read the polarisation results (including likelihoods) back into the sgkit file, from where (I guess) tsinfer will, in the future, be able to access them.

Alternatively, Peter Keightley's method could be reimplemented on top of sgkit, but I think that's probably a lot of work (although maybe suitable for an undergrad/masters project?)

[hyanwong]

One issue is that the best approach may change over time, as more methods are published (and if we even manage to figure out how to improve AA inference via tsinfer).

For this reason, I think it's probably best for the moment to open a GitHub Q&A discussion which points out the best methods, and in the tutorial we can simply direct people to that discussion. Then if there are any improved methods, or user feedback about what works well, we can add it to the Q&A. I'm just sorting this out now. The Q&A is at #523