search

lonelyjoeparker / befi-bats-gui

Java software to analyse phylogeny-trait correlations in discrete traits, accounting for phylogenetic uncertainty via posterior sampling.
http://science.lonelyjoeparker.com
GNU General Public License v3.0
8 stars 3 forks source link

PS scores not consistent between versions #16

Open lonelyjoeparker opened 7 years ago

lonelyjoeparker commented 7 years ago

Some suggestion that the PS implementation changed somewhere 0.7.2 -> 0.10.1 due to generalisation to N states. In particular, the fitch state set pruning algorithm may be implemented differently.

Testing this:

Input (see https://github.com/lonelyjoeparker/befi-bats-gui/tree/master/examples for input files) PS output 0.7.2 PS output 0.10.1
example.trees, (PS = ?) 18.5 25.733...
simulation100.trees, (PS = 4) 4 4
fully_balanced_16tips_2states.trees, (PS = 8) 8 8
fully_balanced_16tips_4states.trees, (PS = 12) 12 12
fully_balanced_16tips_4states-partiallyAgg.trees, (PS = 9...?) 9 10
fully_balanced_8tips_4states-partiallyAgg.trees, (PS = 5...?) 5 5
fully_balanced_16tips_4states-moreAgg.trees (PS=7) 7 9

Basically seems to be overcounting, but in the newer(?) one. assert verbose and inspect states...

NB stast for AI and MC still consistent.