Idea for a fast, CLV-based network pseudolikelihood

[lutteropp]

CLV of a node ->Weighted average of CLVs from different cases. See picture.

[lutteropp]

I can implement that thing in an hour. It is super straight-forward.

[lutteropp]

I implemented it (took about 1.5 hours to code). Turns out it works pretty well! :-)

[lutteropp]

Here I printed the pseudologlikelihood and pseudo-BIC after each accepted move, to show how well this estimation performs:
example-run.txt

Cool thing is that computing this pseudologlikelihood doesn't take much more than computing the loglikelihood of a single phylogenetic tree.

[lutteropp]

My idea is to use pseudologlikelihood during initial prefiltering phase, and then move the top 10% of prefiltered candidates to the normal candidate ranking which uses real loglikelihood for scoring (the scoring involves brlenopt).

[stamatak]

that sounds like a reasonable approach

…

On 30.04.21 14:29, Sarah Lutteropp wrote: My idea is to use pseudologlikelihood during initial prefiltering phase, and then move the top 10% of prefiltered candidates to the normal candidate ranking which uses real loglikelihood for scoring (the scoring involves brlenopt). — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#60 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGXB6XGFDCZ34MQVB5ZHYLTLKIAPANCNFSM432RQ5ZA>.

-- Alexandros (Alexis) Stamatakis Research Group Leader, Heidelberg Institute for Theoretical Studies Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology www.exelixis-lab.org