Just a place to jot down some suggestions, but I realise that the polytomy resolution thing is very much like the normal methods one uses to resolve phylogenetic trees - I could imagine using e.g. an NJ algorithm or a heuristic parsimony search where the thing that you are trying to minimise is number of postulated recombination events. There is presumably also some sort of likelihood method that is the 'correct' one to use.
The issue is, of course, that as the number of children in the polytomy increases, we hit all the same sort of combinatorial problems one hits when normally resolving trees. It would be a pity if the semi-linear scaling of the tsinfer algorithm was scuppered by a ridiculously intensive polytomy resolution step.
Copied from original issue: mcveanlab/treeseq-inference#38
From @hyanwong on February 27, 2017 20:58
Just a place to jot down some suggestions, but I realise that the polytomy resolution thing is very much like the normal methods one uses to resolve phylogenetic trees - I could imagine using e.g. an NJ algorithm or a heuristic parsimony search where the thing that you are trying to minimise is number of postulated recombination events. There is presumably also some sort of likelihood method that is the 'correct' one to use.
The issue is, of course, that as the number of children in the polytomy increases, we hit all the same sort of combinatorial problems one hits when normally resolving trees. It would be a pity if the semi-linear scaling of the tsinfer algorithm was scuppered by a ridiculously intensive polytomy resolution step.
Copied from original issue: mcveanlab/treeseq-inference#38