Package edu.cmu.tetrad.search

Class MarkovCheck

java.lang.Object
edu.cmu.tetrad.search.MarkovCheck
public class MarkovCheck extends Object

Checks whether a graph is locally Markov or locally Faithful given a data set. First a lists of m-separation predictions are made for each pair of variables in the graph given the parents of one of the variables, one list (for local Markov) where the m-separation holds and another list (for local Faithfulness) where the m-separation does not hold. Then the predictions are tested against the data set using the independence test. For the Markov test, since an independence test yielding p-values should be Uniform under the null hypothesis, these p-values are tested for Uniformity using the Kolmogorov-Smirnov test. Also, a fraction of dependent judgments is returned, which should equal the alpha level of the independence test if the test is Uniform under the null hypothesis. For the Faithfulness test, the p-values are tested for Uniformity using the Kolmogorov-Smirnov test; these should be dependent. Also, a fraction of dependent judgments is returned, which should be maximal./p>

A "Markov adequacy score" is also given, which simply returns zero if the Markov p-value Uniformity test fails and the fraction of dependent judgments for the local Faithfulness check otherwise. Maximizing this score picks out models for which Markov holds and faithfulness holds to the extend possible; these model should generally have good accuracy scores.

Author:
josephramsey

  • Constructor Details

    • MarkovCheck

      public MarkovCheck(Graph graph, IndependenceTest independenceTest, MarkovCheck.ConditioningSetType setType)
      Constructor. Takes a graph and an independence test over the variables of the graph.
      Parameters:
      graph - The graph.
      independenceTest - The test over the variables of the graph.

  • Method Details

    • generateResults

      public void generateResults()
      Generates all results, for both the local Markov and local Faithfulness checks, for each node in the graph given the parents of that node. These results are stored in the resultsIndep and resultsDep lists.
      See Also:

    • getAllSubsetsIndependenceFacts

      @NotNull public static @NotNull MarkovCheck.AllSubsetsIndependenceFacts getAllSubsetsIndependenceFacts(Graph graph)

    • getSetType

      public MarkovCheck.ConditioningSetType getSetType()
      Returns type of conditioning sets to use in the Markov check.
      Returns:
      The type of conditioning sets to use in the Markov check.
      See Also:

    • setSetType

      public void setSetType(MarkovCheck.ConditioningSetType setType)
      Sets the type of conditioning sets to use in the Markov check.
      Parameters:
      setType - The type of conditioning sets to use in the Markov check.
      See Also:

    • setParallelized

      public void setParallelized(boolean parallelized)
      True if the checks should be parallelized. (Not always a good idea.)
      Parameters:
      parallelized - True if the checks should be parallelized.

    • getResults

      public List<IndependenceResult> getResults(boolean indep)
      After the generateResults method has been called, this method returns the results for the local Markov or local Faithfulness check, depending on the value of the indep parameter.
      Parameters:
      indep - True for the local Markov results, false for the local Faithfulness results.
      Returns:
      The results for the local Markov or local Faithfulness check.

    • getPValues

      public List<Double> getPValues(List<IndependenceResult> results)
      Returns the list of p-values for the given list of results.
      Parameters:
      results - The results.
      Returns:
      Their p-values.

    • getFractionDependent

      public double getFractionDependent(boolean indep)
      Returns the fraction of dependent judgments for the given list of results.
      Parameters:
      indep - True for the local Markov results, false for the local Faithfulness results.
      Returns:
      The fraction of dependent judgments for this condition.

    • getKsPValue

      public double getKsPValue(boolean indep)
      Returns the Kolmorogov-Smirnov p-value for the given list of results.
      Parameters:
      indep - True for the local Markov results, false for the local Faithfulness results.
      Returns:
      The Kolmorogov-Smirnov p-value for this condition.

    • getMarkovAdequacyScore

      public double getMarkovAdequacyScore(double alpha)
      Returns the Markov Adequacy Score for the graph. This is zero if the p-value of the KS test of Uniformity is less than alpha, and the fraction of dependent pairs otherwise. This is only for continuous Gaussian data, as it hard-codes the Fisher Z test for the local Markov and Faithfulness check.
      Parameters:
      alpha - The alpha level for the KS test of Uniformity. An alpha level greater than this will be considered uniform.
      Returns:
      The Markov Adequacy Score for this graph given the data.

    • getVariables

      public List<Node> getVariables()
      Returns the variables of the independence test.
      Returns:
      The variables of the independence test.

    • getVariable

      public Node getVariable(String name)
      Returns the variable with the given name.
      Parameters:
      name - The name of the variables.
      Returns:
      The variable with the given name.

    • getIndependenceTest

      public IndependenceTest getIndependenceTest()
      Returns the independence test being used.
      Returns:
      This test.