blog
Class MHSampler

java.lang.Object
  blog.Sampler
      blog.MHSampler

Direct Known Subclasses:

DMHSampler

public class MHSampler
extends Sampler

Generates samples from a Markov chain over possible worlds using a Metropolis-Hastings algorithm. The proposal distribution for this algorithm is represented by an object that implements the Proposer interface.

The MHSampler constructor looks at the following properties in the properties table that is passed in:

proposerClass

Name of the proposer class to use. This class must implement the Proposer interface. Default: blog.GenericProposer.

The property table is also passed to the proposer's constructor.

Field Summary

protected Timer	acceptProbTimer
protected PartialWorldDiff	curWorld
protected int	numAcceptedThisTrial
protected int	numSamplesThisTrial
protected int	numTrials
protected Proposer	proposer
protected int	totalNumAccepted
protected int	totalNumSamples
protected Timer	worldUpdateTimer

Fields inherited from class blog.Sampler

evidence, model, queries

Constructor Summary

MHSampler(Model model, java.util.Properties properties)
Creates a new sampler for the given BLOG model.

Method Summary

protected void	constructProposer(java.util.Properties properties) Method responsible for initializing the proposer field.
PartialWorld	getLatestWorld() Returns the world generated by the most recent call to nextSample.
void	initialize(Evidence evidence, java.util.List queries) Prepares this sampler to sample from the distribution conditioned on the given evidence, returning PartialWorld objects that are complete enough to answer the given queries.
void	nextSample() Generates the next world in the Markov chain.
PartialWorld	nextSample(PartialWorld world) Samples from some given world, leaving the current world in sampler undisturbed.
void	printStats() Print statistics gathered during sampling to standard out.
void	setWorld(PartialWorld w)

Methods inherited from class blog.Sampler

getLatestWeight

Methods inherited from class java.lang.Object

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

Field Detail

acceptProbTimer

protected Timer acceptProbTimer

worldUpdateTimer

protected Timer worldUpdateTimer

proposer

protected Proposer proposer

curWorld

protected PartialWorldDiff curWorld

numTrials

protected int numTrials

totalNumSamples

protected int totalNumSamples

totalNumAccepted

protected int totalNumAccepted

numSamplesThisTrial

protected int numSamplesThisTrial

numAcceptedThisTrial

protected int numAcceptedThisTrial

Constructor Detail

MHSampler

public MHSampler(Model model,
                 java.util.Properties properties)

Creates a new sampler for the given BLOG model. The properties table specifies configuration parameters.

Method Detail

constructProposer

protected void constructProposer(java.util.Properties properties)

Method responsible for initializing the proposer field.

initialize

public void initialize(Evidence evidence,
                       java.util.List queries)

Description copied from class: Sampler

Prepares this sampler to sample from the distribution conditioned on the given evidence, returning PartialWorld objects that are complete enough to answer the given queries. Also clears the internal state of this sampler so that the next sample generated will be independent of all previous ones.

The default implementation just sets the evidence and queries member variables.

Overrides:

initialize in class Sampler

queries - List of Query objects

setWorld

public void setWorld(PartialWorld w)

nextSample

public void nextSample()

Generates the next world in the Markov chain. This method gets a proposed world from the Proposer. If the proposal is accepted, then the next world is the proposed world; otherwise the next world is the same as the previous world.

Specified by:

nextSample in class Sampler

nextSample

public PartialWorld nextSample(PartialWorld world)

Samples from some given world, leaving the current world in sampler undisturbed.

getLatestWorld

public PartialWorld getLatestWorld()

Description copied from class: Sampler

Returns the world generated by the most recent call to nextSample. The returned PartialWorld object may be modified by the next call to nextSample.

Specified by:

getLatestWorld in class Sampler

printStats

public void printStats()

Print statistics gathered during sampling to standard out. These figures are gathered during each call to sample() and is called once at the end of each trial.

Overrides:

printStats in class Sampler