de.up.ling.irtg.sampling

Class TreeSample<Type>

java.lang.Object

de.up.ling.irtg.sampling.TreeSample<Type>

Type Parameters:

Type -

public class TreeSample<Type>
extends Object

Represents all the trees drawn in a round of importance sampling along with the necessary weights. Note that all values need to be set externally.

Author:

christoph

Constructor Summary

Constructors

Constructor and Description
TreeSample()

Method Summary

Modifier and Type	Method and Description
void	addSample(Tree<Type> sample) Adds a sample with all the weights set to 0.0.
void	clear() Clears out all the values stores in this sample.
void	expoNormalize(boolean deterministic) Sets self normalized weight for each entry to the appropriate self normalized value for each entry.
void	flatten(org.apache.commons.math3.random.RandomGenerator rg, int size, boolean deterministic) Resamples the trees so that they all have uniform weight.
double	getLogPropWeight(int entry) Returns the log of the proposal probability for the given entry.
double	getLogSumWeight(int entry) Returns the value of the log of the sum of the proposal probabilities for the given entry.
double	getLogTargetWeight(int entry) Returns the log of the target weight for the given entry.
Tree<Type>	getSample(int number) Returns the sample in the given position.
double	getSelfNormalizedWeight(int entry) Returns the weight derived by the different self normalization approaches.
double	makeMaxBase(boolean deterministic, double originalBase) Computes a 'self normalized' value for adaption.
int	populationSize() Returns the size of the underlying population.
void	resample(org.apache.commons.math3.random.RandomGenerator rg, int size) Resamples the trees assuming that getSelfNormalized weight already returns the correct values.
void	resampleWithNormalize(org.apache.commons.math3.random.RandomGenerator rg, int size, boolean deterministic) Resamples the trees.
void	setLogPropWeight(int entry, double amount) Sets the log of the proposal probability for the given entry.
void	setLogSumWeight(int entry, double amount) Sets the value of the log of the sum of the proposal probabilities for the given entry.
void	setLogTargetWeight(int entry, double amount) Sets the log of the target weight for the given entry.
String	toString()

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

TreeSample

public TreeSample()

Method Detail

addSample

public void addSample(Tree<Type> sample)

Adds a sample with all the weights set to 0.0.

Parameters:

sample -

setLogSumWeight

public void setLogSumWeight(int entry,
                            double amount)

Sets the value of the log of the sum of the proposal probabilities for the given entry.

Parameters:

entry -

amount -

setLogPropWeight

public void setLogPropWeight(int entry,
                             double amount)

Sets the log of the proposal probability for the given entry.

Parameters:

entry -

amount -

setLogTargetWeight

public void setLogTargetWeight(int entry,
                               double amount)

Sets the log of the target weight for the given entry.

Parameters:

entry -

amount -

getLogPropWeight

public double getLogPropWeight(int entry)

Returns the log of the proposal probability for the given entry.

Parameters:

entry -

Returns:

getLogSumWeight

public double getLogSumWeight(int entry)

Returns the value of the log of the sum of the proposal probabilities for the given entry.

Parameters:

entry -

Returns:

getLogTargetWeight

public double getLogTargetWeight(int entry)

Returns the log of the target weight for the given entry.

Parameters:

entry -

Returns:

getSelfNormalizedWeight

public double getSelfNormalizedWeight(int entry)

Returns the weight derived by the different self normalization approaches.

Parameters:

entry -

Returns:

getSample

public Tree<Type> getSample(int number)

Returns the sample in the given position.

Parameters:

number -

Returns:

expoNormalize

public void expoNormalize(boolean deterministic)

Sets self normalized weight for each entry to the appropriate self normalized value for each entry. Deterministic decides whether we use the log proposal weight or the log sum of proposals in order to compute the unnormalized importance weight.

Parameters:

deterministic -

populationSize

public int populationSize()

Returns the size of the underlying population.

Returns:

flatten

public void flatten(org.apache.commons.math3.random.RandomGenerator rg,
                    int size,
                    boolean deterministic)

Resamples the trees so that they all have uniform weight. First applies resampleWithNormalize.

Parameters:

rg -

size -

deterministic -

resampleWithNormalize

public void resampleWithNormalize(org.apache.commons.math3.random.RandomGenerator rg,
                                  int size,
                                  boolean deterministic)

Resamples the trees. Note that this makes the weight settings (except for self normalized) invalid.

Parameters:

rg -

size -

deterministic -

resample

public void resample(org.apache.commons.math3.random.RandomGenerator rg,
                     int size)

Resamples the trees assuming that getSelfNormalized weight already returns the correct values.

Parameters:

rg -

size -

toString

public String toString()

Overrides:

toString in class Object

makeMaxBase

public double makeMaxBase(boolean deterministic,
                          double originalBase)

Computes a 'self normalized' value for adaption. Here we do not need the values to sum to 1.0, so we simply use a maximum, which is the maximum of original base and the largest importance weight in the current sequence, to 'normalize' all importance weights. This is intended to prevent underflow before we exponentiate. The resulting importance weights are the accessible by getSelfNormalizedWeight.

Parameters:

deterministic -

originalBase -

Returns:

clear

public void clear()

Clears out all the values stores in this sample.