gov.sandia.cognition.statistics

Class TransferEntropy

java.lang.Object

gov.sandia.cognition.statistics.TransferEntropy

@PublicationReference(author="Thomas Schreiber",
                      title="Measuring Information Transfer",
                      type=Journal,
                      year=2000,
                      url="http://journals.aps.org/prl/pdf/10.1103/PhysRevLett.85.461")
public class TransferEntropy
extends java.lang.Object

A class for calculating the transfer entropy of two vectors. Note that in the current implementation only values included in a transfer entropy object are used in calculating the conditional probability. For example: assume the two vectors were [0,0,0,0,1] and [1,1,1,1,1] with window size 1. Then, when computing the conditional probabilities, the state [1,1] will not be included in the calculations since there's no triad that includes [1,1] in the prior. This generates more intuitive results (the above example has a transfer entropy value of 0 in this implementation), and is easier to code.

Since:

3.4.2

Author:

trbroun

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	TransferEntropy.TransferEntropyDistributionObject A helper class that define the objects used by the distributions in transfer entropy.
static class	TransferEntropy.TransferEntropyPartialSumObject Helper class for holding information about the partial sums.

Constructor Summary

Constructors

Constructor and Description
TransferEntropy(DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> oneTermPriorDistribution, DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> oneTermPosteriorDistribution, DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> twoTermPriorDistribution, DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> twoTermPosteriorDistribution) Creates a new transfer entropy object using distributions.
TransferEntropy(java.util.List<? extends java.lang.Object> sourceList, java.util.List<? extends java.lang.Object> destinationList, int windowSize) Creates a new transfer entropy object using lists instead of vectors.
TransferEntropy(java.util.List<? extends java.lang.Object> sourceList, java.util.List<? extends java.lang.Object> destinationList, int sourceWindowSize, int destinationWindowSize) Creates a new transfer entropy object using lists instead of vectors.
TransferEntropy(Vector sourceVector, Vector destinationVector, int windowSize) Creates a new transfer entropy object
TransferEntropy(Vector sourceVector, Vector destinationVector, int sourceWindowSize, int destinationWindowSize) Creates a new transfer entropy object.

Method Summary

Modifier and Type	Method and Description
double	compute() Computes the transfer entropy value.
Pair<java.lang.Double,java.util.HashMap<TransferEntropy.TransferEntropyDistributionObject,TransferEntropy.TransferEntropyPartialSumObject>>	computeWithPartialSums() Computes the transfer entropy value and returns a map of the states with their corresponding partial sums.

Methods inherited from class java.lang.Object

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

Constructor Detail

TransferEntropy

public TransferEntropy(Vector sourceVector,
                       Vector destinationVector,
                       int windowSize)

Creates a new transfer entropy object

Parameters:

destinationVector -

sourceVector -

windowSize - How many steps in the past will be used in the analysis.

TransferEntropy

public TransferEntropy(Vector sourceVector,
                       Vector destinationVector,
                       int sourceWindowSize,
                       int destinationWindowSize)

Creates a new transfer entropy object.

Parameters:

destinationVector -

sourceVector -

destinationWindowSize - How many time steps into the past will be used with the vectorBeingActedOn

sourceWindowSize - How many time steps into the past will be sued with the vectorActingOnOtherVector

TransferEntropy

public TransferEntropy(java.util.List<? extends java.lang.Object> sourceList,
                       java.util.List<? extends java.lang.Object> destinationList,
                       int windowSize)

Creates a new transfer entropy object using lists instead of vectors.

Parameters:

destinationList -

sourceList -

windowSize - How many steps in the past will be used in the analysis.

TransferEntropy

public TransferEntropy(java.util.List<? extends java.lang.Object> sourceList,
                       java.util.List<? extends java.lang.Object> destinationList,
                       int sourceWindowSize,
                       int destinationWindowSize)

Creates a new transfer entropy object using lists instead of vectors.

Parameters:

destinationList -

sourceList -

destinationWindowSize - How many time steps into the past will be used with the vectorBeingActedOn

sourceWindowSize - How many time steps into the past will be sued with the vectorActingOnOtherVector

TransferEntropy

public TransferEntropy(DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> oneTermPriorDistribution,
                       DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> oneTermPosteriorDistribution,
                       DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> twoTermPriorDistribution,
                       DefaultDataDistribution<TransferEntropy.TransferEntropyDistributionObject> twoTermPosteriorDistribution)

Creates a new transfer entropy object using distributions.

Parameters:

oneTermPriorDistribution - The prior distribution over one vector. The prior one term prior should be filled in, while the second vector prior and posterior should be null (or a consistent value over each element of the domain).

oneTermPosteriorDistribution - A distribution over the first term prior and the posterior. Leave the second term vector prior as null, or a consistent value across the domain space.

twoTermPriorDistribution - A distribution over TransferEntropyDistributionObjects where the prior for the first and second vector is supplied.

twoTermPosteriorDistribution - A distribution over TransferEntropyDistributionObjects where all values are supplied.

Method Detail

compute

public double compute()

Computes the transfer entropy value.

Returns:

The transfer entropy from one vector to the other, or one distribution to the other.

computeWithPartialSums

public Pair<java.lang.Double,java.util.HashMap<TransferEntropy.TransferEntropyDistributionObject,TransferEntropy.TransferEntropyPartialSumObject>> computeWithPartialSums()

Computes the transfer entropy value and returns a map of the states with their corresponding partial sums. Used to identify which states contribute the most or least to the transfer entropy.

Returns:

A pair, where the first element is the transfer entropy value between the two vectors, and the second element is a map from states to their corresponding partial sums.