[Help Wanted] Running MCMC under the prior also removes the node calibrations

[bjoelle]

Running the mcmc.run with underPrior = TRUE seems to remove all clamped nodes, including node calibrations (at least if node calibrations are set following this tutorial here)
I don't think this is proper behaviour as node calibrations are generally considered part of the prior - in particular this makes difficult to check the effective priors on the calibrations, which is generally good practice

[hoehna]

Yes, I had such an example recently too. This is a bit tricky given the way how node calibrations are currently setup in RevBayes.

To get around the problem, this is what you can try:

1. Use xxx.setValue( 0.0 ) instead of xxx.clamp( 0.0 ) for the node calibrations, or
2. Remove the dnPhyloCTMC and run the MCMC as usual.

I did the second option.

[bjoelle]

thanks, that is helpful to know !

[bredelings]

Yeah, "remove the dnPhyloCTMC and run the MCMC as usual" seems like it corresponds more closely to what BEAST people mean by "run under the prior". What they really mean is "run without the sequence data", which is exactly what happens when you remove dnPhyloCTMC( ).