Sarah Lutteropp
Sarah Lutteropp
@stamatak Already in there ;-) I copy-pasted the struct I use for the results, as this is the shortest code part where one can see what is currently evaluated ```...
This is what we currently measure: https://github.com/lutteropp/NetRAX/issues/14
I just realized that we may need both PERFECT_SAMPLING and PERFECT_UNIFORM_SAMPLING in our experiments: - PERFECT_SAMPLING is better for comparing loglikelihood and BIC score. - PERFECT_UNIFORM_SAMPLING is better for scoring...
There are three different settings on how to handle partitions: - PLLMOD_COMMON_BRLEN_UNLINKED: Each partition has its own independent branch lengths and reticulation probs. - PLLMOD_COMMON_BRLEN_LINKED: All partitions share the same...
@stamatak umm... all 3 versions are fully implemented in NetRAX already
should I switch the experiments to run with UNLINKED then? So far I used LINKED.
Cool, then let's stick to _scaled_ (and if we under-estimate reticulations, switch to linked later) and only _PERFECT_SAMPLING_ first :) It is easier (in terms of total runtime, number of...
Yes, let's make a separate experiment that only looks at the effect of scaled vs. unlinked brlens :) EDIT: I created an issue for it: https://github.com/lutteropp/NetRAX/issues/28
(I can do the runs with tons of different parameters on the cluster, but my problem is interpreting, plotting, and summarizing all these results for so many different parameter combinations...
(also, having to entirely re-run some broken experiment is more painful when it was on a lot of parameter combinations)