Class MultivariateNormalMixtureExpectationMaximization

java.lang.Object
org.hipparchus.stat.fitting.MultivariateNormalMixtureExpectationMaximization

public class MultivariateNormalMixtureExpectationMaximization extends Object
Expectation-Maximization algorithm for fitting the parameters of multivariate normal mixture model distributions. This implementation is pure original code based on EM Demystified: An Expectation-Maximization Tutorial by Yihua Chen and Maya R. Gupta, Department of Electrical Engineering, University of Washington, Seattle, WA 98195. It was verified using external tools like CRAN Mixtools (see the JUnit test cases) but it is not based on Mixtools code at all. The discussion of the origin of this class can be seen in the comments of the MATH-817 JIRA issue.
  • Constructor Details

  • Method Details

    • fit

      public void fit(MixtureMultivariateNormalDistribution initialMixture, int maxIterations, double threshold) throws MathIllegalArgumentException
      Fit a mixture model to the data supplied to the constructor. The quality of the fit depends on the concavity of the data provided to the constructor and the initial mixture provided to this function. If the data has many local optima, multiple runs of the fitting function with different initial mixtures may be required to find the optimal solution. If a MathIllegalArgumentException is encountered, it is possible that another initialization would work.
      Parameters:
      initialMixture - Model containing initial values of weights and multivariate normals
      maxIterations - Maximum iterations allowed for fit
      threshold - Convergence threshold computed as difference in logLikelihoods between successive iterations
      Throws:
      MathIllegalArgumentException - if any component's covariance matrix is singular during fitting
      MathIllegalArgumentException - if numComponents is less than one or threshold is less than Double.MIN_VALUE
      MathIllegalArgumentException - if initialMixture mean vector and data number of columns are not equal
    • fit

      Fit a mixture model to the data supplied to the constructor. The quality of the fit depends on the concavity of the data provided to the constructor and the initial mixture provided to this function. If the data has many local optima, multiple runs of the fitting function with different initial mixtures may be required to find the optimal solution. If a MathIllegalArgumentException is encountered, it is possible that another initialization would work.
      Parameters:
      initialMixture - Model containing initial values of weights and multivariate normals
      Throws:
      MathIllegalArgumentException - if any component's covariance matrix is singular during fitting
      MathIllegalArgumentException - if numComponents is less than one or threshold is less than Double.MIN_VALUE
    • estimate

      public static MixtureMultivariateNormalDistribution estimate(double[][] data, int numComponents) throws MathIllegalArgumentException
      Helper method to create a multivariate normal mixture model which can be used to initialize fit(MixtureMultivariateNormalDistribution). This method uses the data supplied to the constructor to try to determine a good mixture model at which to start the fit, but it is not guaranteed to supply a model which will find the optimal solution or even converge.
      Parameters:
      data - Data to estimate distribution
      numComponents - Number of components for estimated mixture
      Returns:
      Multivariate normal mixture model estimated from the data
      Throws:
      MathIllegalArgumentException - if numComponents is greater than the number of data rows.
      MathIllegalArgumentException - if numComponents < 2.
      MathIllegalArgumentException - if data has less than 2 rows
      MathIllegalArgumentException - if rows of data have different numbers of columns
    • getLogLikelihood

      public double getLogLikelihood()
      Gets the log likelihood of the data under the fitted model.
      Returns:
      Log likelihood of data or zero of no data has been fit
    • getFittedModel

      public MixtureMultivariateNormalDistribution getFittedModel()
      Gets the fitted model.
      Returns:
      fitted model or null if no fit has been performed yet.