For how long do we have to run the chains?

matsengrp / ecgtheow

Ancestral lineage reconstruction using BEAST or RevBayes

2 stars 2 forks source link

For how long do we have to run the chains? #9

Closed matsen closed 6 years ago

matsen commented 6 years ago

What sort of convergence should we be seeing in terms of ESS of various parameters in order to think that the ancestral lineage graph is stable (i.e. it won't change with longer MCMC runs)?

lauradoepker commented 6 years ago

I ran the 125-pruned version of our data for 500 million iterations of BEAST. I have all the resultant files, but @dunleavy005, do I just use tracer to generate the caterpillar graphic you made a while ago? I can do that for the 500 million run. Our default setting is 10 million, so I can run tracer on all the ecgtheow runs I'm about to do with the new partis runs Duncan's done.

dunleavy005 commented 6 years ago

Yes, use tracer, it has a nice GUI so it should spit out the trace plot almost immediately.

lauradoepker commented 6 years ago

@dunleavy005 uploaded this caterpillar trace plot for the run I did on 500 million iterations. I think we decided it was sufficiently random? @matsen

matsen commented 6 years ago

Oh, yes, it's lovely.

dunleavy005 commented 6 years ago

Can now dynamically tweak mcmc iters from #21