BEAST v2.7.8 Documentation: starbeast3.operators.ParallelMCMCTreeOperator

Proposes a move in state space.
Operator that does proposals that count for one step or more steps in the MCMC
Run MCMC on different gene tree parts of the model in parallel before combining them in a single Gibbs move

Inputs:

bactrian, includeRealParameters, exclude, include, logsum, distribution, unthread, chainLength, chainCoverage, threads, otherState, learning, burnin, schedule, nregression, targetCPU, targetWeight, runtime, speciesTree, weight

 

bactrian
type: java.lang.Boolean
flag to indicate that bactrian operators should be used where possible
Optional input. Default: true

 

includeRealParameters
type: java.lang.Boolean
flag to include Real Parameters for each of the partitions in the analysis
Optional input. Default: true

 

exclude
type: beast.base.inference.StateNode***
parameters to ensure are not operated on
Optional input

 

include
type: beast.base.inference.StateNode***
parameters to add to the state, in case they are not automatically detected
Optional input

 

logsum
type: beast.base.inference.StateNode***
these parameters, if operated on, will have their total sum respected by the avmn operator
Optional input

 

distribution
type: starbeast3.operators.ParallelDistSet***
Distribution on a tree conditinionally independent from all other distributions given the state of the restof parameter space.
Optional input

 

unthread
type: java.lang.Boolean
flag to convert ThreadedTreeLikelihood back into TreeLikelihood when this is called
Optional input. Default: false

 

chainLength
type: java.lang.Long
Length of the MCMC chain: each individual ParallelMCMC performs chainLength/nrOfThreads samples
Either this, or chainCoverage needs to be specified

 

chainCoverage
type: java.lang.Double
The MCMC chain length is the coverage times the number of parameters
Either this, or chainLength needs to be specified

 

threads
type: java.lang.Integer
maximum number of threads to use, if less than 1 the number of threads in BeastMCMC is used (default -1)
Optional input. Default: -1

 

otherState
type: beast.base.inference.State
main state containing all statenodes for this analysis
Optional input

 

learning
type: java.lang.Boolean
Learn whether to parallelise (n threads) or not (1 thread 1 operator)
Optional input. Default: true

 

burnin
type: java.lang.Integer
How many operator calls before thread learning kicks in. Learning will begin after chainLength regression.
Optional input. Default: 10000

 

schedule
type: starbeast3.core.OperatorScheduleRecalculator
Operator schedule (if learning is applied)
Optional input

 

nregression
type: java.lang.Integer
Number of MCMC chainLengths vs runtimes to sample in order to learn chainLengths, for load balancing. Set to <5 to skip the training.
Optional input. Default: 200

 

targetCPU
type: java.lang.Double
Proportion of threads allocated (if > 1) that should be spent on MCMC, as opposed to overhead.Larger targetCPU will mean longer chains and lower operator weight. If targetCPU=0, then the load balancing will match the slowest thread. Set nregression=0 to omit this step.
Optional input. Default: 0.8

 

targetWeight
type: java.lang.Double
Target effective weight of this operator, to be learned if regression is applied. The effective weight is the operator weight * chainLength sum. Set this to 0 (or nregression=0) to omit this step.
Optional input. Default: 0.0

 

runtime
type: java.lang.Double
Max runtime of MCMC chains during training (only applicable if load balancing is being trained). If this isset to -1, then chain lengths are sampled instead of runtimes.
Optional input. Default: -1.0

 

speciesTree
type: beast.base.evolution.tree.Tree
an optional dummy input so that beauti can load the template (hack)
Optional input

 

weight
type: java.lang.Double
weight with which this operator is selected
Required input