Welcome, Guest. Please Login
YaBB - Yet another Bulletin Board
Optimisation issue (Read 2073 times)
Miguel Lacerda
YaBB Newbies

Feed your monkey!

Posts: 36
Natl Univ of Ireland, Galway
Gender: male
Optimisation issue
Oct 3rd, 2013 at 12:25am
Hi Sergei,

I am having issues with the optimisation of a GY94+GDD3 model on a partitioned alignment (separate trees in each of two partitions). When I optimise the model on silverback, I get the following error repeated several times at the end of the messages.log file:

Internal error in _LikelihoodFunction::ConjugateGradientDescent. The function evaluated at current parameter values [-3171489.841306037] does not match the last recorded LF maximum [-9167055.605034702]

I do not get this message when I run the same code on a different cluster [running HYPHY 2.1120130213beta(MPI)], but the final parameter estimates still don't look right; all the branch lengths on the one tree are set to ~0.1.

I have subsequently tried to optimise the model parameters on each partition separately; however, the likelihood of the final parameter estimates (-5 082 591) is worse than the likelihood based on the initial parameter values (-3 084 260).

I cannot upload my data and code here as it apparently "exceeds the maximum allowed upload-directory volume" (despite being only 86kB!), so I will send it to you via email. I have 818 codon sites (637 variable sites) and I am trying to estimate 900 parameters over both trees. Could the overparameterisation be the issue?

Back to top
IP Logged
YaBB Administrator

Datamonkeys are forever...

Posts: 1658
Gender: male
Re: Optimisation issue
Reply #1 - Oct 9th, 2013 at 11:44am
Hi Miguel,

I am looking into the issue.

Back to top

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
IP Logged