org.broadinstitute.hellbender.utils.mcmc

Class GibbsSampler<V extends java.lang.Enum<V> & ParameterEnum,S extends ParameterizedState<V>,T extends DataCollection>

java.lang.Object

org.broadinstitute.hellbender.utils.mcmc.GibbsSampler<V,S,T>

public final class GibbsSampler<V extends java.lang.Enum<V> & ParameterEnum,S extends ParameterizedState<V>,T extends DataCollection>
extends java.lang.Object

Implements Gibbs sampling of a multivariate probability density function. See GibbsSamplerSingleGaussianUnitTest and GibbsSamplerCopyRatioUnitTest for examples of use.

Constructor Summary

Constructors

Constructor and Description
GibbsSampler(int numSamples, ParameterizedModel<V,S,T> model) Constructs a GibbsSampler given the total number of samples (including burn-in) and a ParameterizedModel.

Method Summary

Modifier and Type	Method and Description
<U> java.util.List<U>	getSamples(V parameterName, java.lang.Class<U> parameterValueClass, int numBurnIn) Returns a list of samples for a specified model parameter, discarding the first numBurnIn samples.
void	runMCMC() Runs the Monte Carlo Markov Chain, using the state of the model provided in the constructor to initialize.
void	setNumSamplesPerLogEntry(int numSamplesPerLogEntry) Changes the number of samples per log entry.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

GibbsSampler

public GibbsSampler(int numSamples,
                    ParameterizedModel<V,S,T> model)

Constructs a GibbsSampler given the total number of samples (including burn-in) and a ParameterizedModel. The ParameterizedState held by the model is used to initialize the Monte Carlo Markov Chain and is taken to be the first sample. Number of samples per log entry will be set to the default.

Parameters:

numSamples - total number of samples; must be positive

model - ParameterizedModel to be sampled

Method Detail

setNumSamplesPerLogEntry

public void setNumSamplesPerLogEntry(int numSamplesPerLogEntry)

Changes the number of samples per log entry.

Parameters:

numSamplesPerLogEntry - number of samples per log entry; must be positive

runMCMC

public void runMCMC()

Runs the Monte Carlo Markov Chain, using the state of the model provided in the constructor to initialize. Progress is logged according to numSamplesPerLogEntry.

getSamples

public <U> java.util.List<U> getSamples(V parameterName,
                                        java.lang.Class<U> parameterValueClass,
                                        int numBurnIn)

Returns a list of samples for a specified model parameter, discarding the first numBurnIn samples.

Type Parameters:

U - type of parameter value

Parameters:

parameterName - name of parameter

parameterValueClass - class of parameter value

numBurnIn - number of burn-in samples to discard from beginning of chain

Returns:

List of parameter samples