Module es.upm.fi.cig.multictbnc
Package es.upm.fi.cig.multictbnc.learning.parameters.ctbn

Class CTBNSufficientStatistics

java.lang.Object
es.upm.fi.cig.multictbnc.learning.parameters.ctbn.CTBNSufficientStatistics
All Implemented Interfaces:
SufficientStatistics
public class CTBNSufficientStatistics extends Object implements SufficientStatistics
Computes and stores the sufficient statistics of a discrete CTBN node. The sufficient statistics are:

(1) Mxy: number of times a variable transitions from a certain state to another one while its parents take a certain instantiation.

(2) Mx: number of times a variable leaves a certain state (for any other state) while its parents take a certain instantiation.

(3) Tx: time that a variable stays in a certain state while its parents take a certain instantiation.

  • Constructor Summary

    Constructors
    Constructor
    Description
    CTBNSufficientStatistics(double mHP, double tHP)
    Receives the hyperparameters of the Dirichlet prior distribution over the parameters that are necessary for Bayesian estimation.

  • Method Summary

    Modifier and Type
    Method
    Description
    void
    computeSufficientStatistics(DiscreteStateNode node, Dataset dataset)
    Computes the sufficient statistics of a CTBN node.
    double[][]
    getMx()
    Returns the sufficient statistic with the number of times the variable leaves every state (i.e., the state changes) while its parents have certain values.
    double
    getMxHyperparameter()
    Returns the value of the hyperparameter with the number of 'imaginary' transitions that occurred from a certain state before seeing the data.
    double[][][]
    getMxy()
    Returns the sufficient statistic with the number of times the variable transition from a certain state to another while its parents have certain values
    double
    getMxyHyperparameter()
    Returns the value of the hyperparameter with the number of 'imaginary' transitions that occurred from a certain state to another before seeing the data.
    double[][]
    getTx()
    Returns the sufficient statistic with the time that the variable stays in every state, while its parents take different values.
    double
    getTxHyperparameter()
    Returns the value of the hyperparameter with the 'imaginary' time that was spent in a certain state before seeing the data.
    protected void
    initialiseSufficientStatistics(DiscreteStateNode node)
    Initialises the structures to store the sufficient statistics of the node.
    protected void
    updateMx(int stateParents, int fromState, double numOccurrences)
    Updates the number of occurrences where the node transitions from a certain state to any other state given an instantiation of its parents.
    protected void
    updateMxy(int stateParents, int fromState, int toState, double numOccurrences)
    Updates the number of occurrences where the node transitions from a certain state to another given an instantiation of its parents.
    protected void
    updateSufficientStatistics(Sequence sequence, int idxFromState, int idxtoState, int idxStateParents, int idxObservation)
    Update the values of the sufficient statistics.

    Methods inherited from class java.lang.Object

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

  • Constructor Details

    • CTBNSufficientStatistics

      public CTBNSufficientStatistics(double mHP, double tHP)
      Receives the hyperparameters of the Dirichlet prior distribution over the parameters that are necessary for Bayesian estimation.
      Parameters:
      mHP - number of times a variable transitions from a certain state (hyperparameter)
      tHP - time that a variable stays in a certain state (hyperparameter)

  • Method Details

    • getMx

      public double[][] getMx()
      Returns the sufficient statistic with the number of times the variable leaves every state (i.e., the state changes) while its parents have certain values.
      Returns:
      number of times the variable leaves every state

    • getMxHyperparameter

      public double getMxHyperparameter()
      Returns the value of the hyperparameter with the number of 'imaginary' transitions that occurred from a certain state before seeing the data.
      Returns:
      hyperparameter with the number of 'imaginary' transitions that occurred from a certain state

    • getMxy

      public double[][][] getMxy()
      Returns the sufficient statistic with the number of times the variable transition from a certain state to another while its parents have certain values
      Returns:
      number of occurrences of every transition

    • getMxyHyperparameter

      public double getMxyHyperparameter()
      Returns the value of the hyperparameter with the number of 'imaginary' transitions that occurred from a certain state to another before seeing the data.
      Returns:
      hyperparameter with the number of 'imaginary' transitions that occurred from a certain state to another

    • getTx

      public double[][] getTx()
      Returns the sufficient statistic with the time that the variable stays in every state, while its parents take different values.
      Returns:
      time that the variable stay for every state

    • getTxHyperparameter

      public double getTxHyperparameter()
      Returns the value of the hyperparameter with the 'imaginary' time that was spent in a certain state before seeing the data.
      Returns:
      hyperparameter with the 'imaginary' time that was spent in a certain state

    • computeSufficientStatistics

      public void computeSufficientStatistics(DiscreteStateNode node, Dataset dataset)
      Computes the sufficient statistics of a CTBN node.
      Specified by:
      computeSufficientStatistics in interface SufficientStatistics
      Parameters:
      dataset - dataset from which sufficient statistics are extracted
      node - node whose sufficient statistics are computed

    • initialiseSufficientStatistics

      protected void initialiseSufficientStatistics(DiscreteStateNode node)
      Initialises the structures to store the sufficient statistics of the node. In the case of Bayesian estimation, the imaginary counts are defined based on the number of states of the node and its parents, so the total number of imaginary samples is not influenced by the cardinality of those nodes.
      Parameters:
      node - node

    • updateMx

      protected void updateMx(int stateParents, int fromState, double numOccurrences) throws NeverSeenStateException
      Updates the number of occurrences where the node transitions from a certain state to any other state given an instantiation of its parents.
      Parameters:
      stateParents - index of the state of the node's parents
      fromState - index of the current state
      numOccurrences - number of occurrences
      Throws:
      NeverSeenStateException - thrown if new node's states appeared after the initialisation of the sufficient statistics

    • updateMxy

      protected void updateMxy(int stateParents, int fromState, int toState, double numOccurrences) throws NeverSeenStateException
      Updates the number of occurrences where the node transitions from a certain state to another given an instantiation of its parents.
      Parameters:
      stateParents - index of the state of the node's parents
      fromState - index of the state from which the transition starts
      toState - index of the state to which the transition ends
      numOccurrences - number of times this transition occurs
      Throws:
      NeverSeenStateException - thrown if new node's states appeared after the initialisation of the sufficient statistics

    • updateSufficientStatistics

      protected void updateSufficientStatistics(Sequence sequence, int idxFromState, int idxtoState, int idxStateParents, int idxObservation) throws NeverSeenStateException
      Update the values of the sufficient statistics.
      Parameters:
      sequence - sequence analysed when updating the sufficient statistics
      idxFromState - index of the state from which the transition used to update the sufficient statistics starts
      idxtoState - index of the state to which the transition used to update the sufficient statistics ends
      idxStateParents - index of the state the node's parents had when the transition occurred.
      idxObservation - index of the observation where the transition ends
      Throws:
      NeverSeenStateException - thrown if new node's states appeared after the initialisation of the sufficient statistics