BEAST2 EBSP warning: "WARNING: no down item specified in UpDownOperator"

138 views
Skip to first unread message

Charles Foster

unread,
Jul 9, 2019, 9:32:37 PM7/9/19
to beast-users
Hi all,

I have a quick question about a warning I get in the BEAST 2.5.2 output when using the Extended Bayesian Skyline Plot tree prior. The following warning is output:

WARNING: no down item specified in UpDownOperator


I looked through the XML file to see which UpDownOperator was missing a down item. It was the following:

        <operator id="EBSPupDownOperator.alltrees" spec="UpDownOperator" scaleFactor="0.7" weight="5.0">
                   
           
<up idref="popSizes.alltrees"/>
                   
           
<up idref="populationMean.alltrees"/>
               
       
</operator>



This operator appears to be missing by default for the EBSP prior- it's missing in the template XML code in the Github repository (https://github.com/CompEvol/beast2/blob/master/templates/TreePriors.xml).

Is this warning something to be concerned about? If so, should I remove the operator, or add in down items for "popSizes.alltrees" and "populationMean.alltrees"? Or are no down items included for a good reason and this warning should be ignored?

Cheers,
Charles

Remco Bouckaert

unread,
Jul 10, 2019, 10:37:24 PM7/10/19
to beast...@googlegroups.com
Hi Charles,

The short answer is that the warning can be ignored. In this case, the up/down operator is used to scale both pop sizes and population mean in the same direction, so the warning is harmless. In general, up/down operators are used to scale things in different directions for highly correlated items, like tree and clock rates: the higher the tree, the lower the clock rate, so if the tree is added as ‘up’, the clock rate should be added as ‘down’ in the operator. Potentially, for your analysis, the operator could benefit from adding trees as up (since higher population sizes mean higher trees) and clock rates as down (as outlined before), but that is something you can explore if you find that the MCMC is badly mixing.

Cheers,

Remco

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To post to this group, send email to beast...@googlegroups.com.
Visit this group at https://groups.google.com/group/beast-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/97e4e303-3192-418b-bd68-d01cff1160c5%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply all
Reply to author
Forward
0 new messages