Package edu.cmu.tetrad.search

Interface IsScore

All Known Implementing Classes:
IsBDeuScore, IsBicScore
public interface IsScore
Instance-Specific BIC (IS-BIC) score for discrete data.

This score adapts the standard Bayesian Information Criterion (BIC) to the instance-specific setting. As with the population version, the likelihood term is based on empirical counts, and the penalty term is proportional to the number of free parameters in the local conditional distribution:

   BIC = log-likelihood – 0.5 * penaltyDiscount * (numParams) * log(N)

where numParams = r_p * (K – 1), with r_p equal to the product of category counts of the parent set and K the number of categories of the child variable.

The instance-specific contribution does not alter the likelihood computation itself. Instead, it is incorporated through a structure prior that rewards or penalizes local modifications (addition, removal, or reversal of parents) relative to the baseline population model. In this way, IS-BIC balances population-wide fit with adjustments that highlight edges most relevant to the chosen test case.

This score is intended for use by search algorithms such as IS-FGES and IS-GFCI, providing a lightweight alternative to IS-BDeu when a BIC-style criterion is preferred.

Usage Notes

  • Currently supports discrete variables only.
  • Continuous data should be discretized before applying this score.
Author:
fattaneh

  • Method Summary

    Modifier and Type
    Method
    Description
    boolean
    determines(List<Node> z, Node y)
    Determines whether the given list of nodes has a specific relationship with the specified node.
    DataSet
    getDataSet()
    Retrieves the dataset used in the scoring calculations.
    int
    getMaxDegree()
    Retrieves the maximum allowable degree for nodes in the current scoring context.
    double
    getSamplePrior()
    Retrieves the prior probability assigned to the sample.
    int
    getSampleSize()
    Retrieves the sample size used in the score computation.
    double
    getStructurePrior()
    Retrieves the prior value assigned to the structure.
    Node
    getVariable(String targetName)
    Retrieves a variable node by its target name.
    List<Node>
    getVariables()
    Retrieves a list of variables represented as nodes.
    boolean
    isEffectEdge(double bump)
    Determines whether the edge has a significant effect based on the given bump value.
    double
    localScore(int node, int[] parents_is, int[] parents_pop, int[] children_population)
    Calculates the local score for a given node in the context of its parent nodes and children population.
    double
    localScore1(int node, int[] parents_is, int[] parents_pop, int[] children_pop)
    Calculates the local score for a given node in the context of its parent nodes and children population, without using structure prior.
    double
    localScoreDiff(int x, int y, int[] parentsY_is, int[] parentsY_pop, int[] childrenY_pop)
    Calculates the difference between local scores before and after introducing a change in the relationship between nodes x and y, considering their parent and children populations.
    void
    setSamplePrior(double samplePrior)
    Sets the prior probability value assigned to the sample.
    void
    setStructurePrior(double structurePrior)
    Sets the prior value assigned to the structure.

  • Method Details

    • localScore

      double localScore(int node, int[] parents_is, int[] parents_pop, int[] children_population)
      Calculates the local score for a given node in the context of its parent nodes and children population.
      Parameters:
      node - The index of the node for which the score is being calculated.
      parents_is - An array representing the indices of the parent nodes.
      parents_pop - An array representing the population of the parent nodes.
      children_population - An array representing the population of the children nodes.
      Returns:
      The local score for the specified node.

    • localScoreDiff

      double localScoreDiff(int x, int y, int[] parentsY_is, int[] parentsY_pop, int[] childrenY_pop)
      Calculates the difference between local scores before and after introducing a change in the relationship between nodes x and y, considering their parent and children populations.
      Parameters:
      x - The index of the node x.
      y - The index of the node y.
      parentsY_is - An array of indices representing the parent nodes of y.
      parentsY_pop - An array of indices representing the population of parent nodes of y.
      childrenY_pop - An array of indices representing the population of children nodes of y.
      Returns:
      The difference in local scores due to the relationship change between x and y.

    • getVariables

      List<Node> getVariables()
      Retrieves a list of variables represented as nodes.
      Returns:
      A list of nodes representing the variables.

    • isEffectEdge

      boolean isEffectEdge(double bump)
      Determines whether the edge has a significant effect based on the given bump value.
      Parameters:
      bump - The threshold value to evaluate the significance of the effect on the edge.
      Returns:
      true if the edge has a significant effect; false otherwise.

    • getSampleSize

      int getSampleSize()
      Retrieves the sample size used in the score computation.
      Returns:
      the sample size.

    • getVariable

      Node getVariable(String targetName)
      Retrieves a variable node by its target name.
      Parameters:
      targetName - the name of the target variable to retrieve
      Returns:
      the node representing the variable with the specified target name

    • getMaxDegree

      int getMaxDegree()
      Retrieves the maximum allowable degree for nodes in the current scoring context.
      Returns:
      The maximum degree a node can have.

    • determines

      boolean determines(List<Node> z, Node y)
      Determines whether the given list of nodes has a specific relationship with the specified node.
      Parameters:
      z - the list of nodes to be evaluated.
      y - the node to be checked against the list.
      Returns:
      true if the list of nodes determines the specified node; false otherwise.

    • localScore1

      double localScore1(int node, int[] parents_is, int[] parents_pop, int[] children_pop)
      Calculates the local score for a given node in the context of its parent nodes and children population, without using structure prior.

      same as localSCire but this one doesn't use structure prior

      Parameters:
      node - The index of the node for which the score is being calculated.
      parents_is - An array representing the indices of the parent nodes.
      parents_pop - An array representing the population of the parent nodes.
      children_pop - An array representing the population of the children nodes.
      Returns:
      The local score for the specified node.

    • getStructurePrior

      double getStructurePrior()
      Retrieves the prior value assigned to the structure.
      Returns:
      the structure prior value as a double.

    • setStructurePrior

      void setStructurePrior(double structurePrior)
      Sets the prior value assigned to the structure.
      Parameters:
      structurePrior - the structure prior value to be set as a double.

    • getSamplePrior

      double getSamplePrior()
      Retrieves the prior probability assigned to the sample.
      Returns:
      the sample prior value as a double.

    • setSamplePrior

      void setSamplePrior(double samplePrior)
      Sets the prior probability value assigned to the sample.
      Parameters:
      samplePrior - the sample prior value to be set as a double.

    • getDataSet

      DataSet getDataSet()
      Retrieves the dataset used in the scoring calculations.
      Returns:
      the dataset used in the scoring calculations.