Best way to fix topology and branch-lengths

[sohrabsa]

Is the following the correct way to stop a variable from being sampled?

To have both the topology and branch-length variables fixed, remove the priors over them, for instance in SimplePhyloModel.java, remove the treePrior and branchLengthHyperParameter factors by removing the following lines:

@DefineFactor
NonClockTreePrior<RateParameterization> treePrior = 
NonClockTreePrior
.on(likelihood.tree);

@DefineFactor
Exponential<Exponential.MeanParameterization> branchLengthHyperPrior = 
Exponential
.on(treePrior.branchDistributionParameters.rate)
.withMean(10.0);

[jewellsean]

Yes, I think that this would be an effective way to stop variables from being sampled. Blang will sample stochastic nodes (which are unobserved) => if you remove the factor def'ns as you have done above those nodes will no longer be stochastic and therefore not sampled.

(NB: I have not verified this in conifer, and I think it would be easier to test if SimplePhyloModel was cleaned out as described in #24)

[zhaottcrystal]

@sohrabsa
yes, you are right, but make sure you provide a treeFile as the fixed tree topology,
you might also wish to remove this line:
private final PrintWriter treeWriter = BriefIO.output(Results.getFileInResultFolder("tree.nwk"));

since your tree topology is already fixed and you do not need to write them into files.