Package dev.nm.stat.random.rng.multivariate.mcmc.metropolis

Class RobustAdaptiveMetropolis

java.lang.Object
dev.nm.stat.random.rng.multivariate.mcmc.metropolis.AbstractMetropolis
dev.nm.stat.random.rng.multivariate.mcmc.metropolis.RobustAdaptiveMetropolis
All Implemented Interfaces:
RandomVectorGenerator, Seedable
public class RobustAdaptiveMetropolis extends AbstractMetropolis
A variation of Metropolis, that uses the estimated covariance of the target distribution in the proposal distribution, based on a paper by Vihola (2011). The algorithm adjusts it's approximation of the variance/covariance after each sample has been drawn, according to a decaying step size, in order to match the given target acceptance rate.
See Also:
    • Vihola, M. "Robust adaptive Metropolis algorithm with coerced acceptance rate," Statistics and Computing. 2011.
    • R package: adaptMCMC

  • Constructor Details

    • RobustAdaptiveMetropolis

      public RobustAdaptiveMetropolis(RealScalarFunction logf, Matrix initialScale, double gamma, double targetAcceptance, Vector initialState, RandomStandardNormalGenerator rnorm, RandomLongGenerator uniform)
      Constructs a new instance with the given parameters.
      Parameters:
      logf - the log of the unnormalized pdf from which we wish to sample
      initialScale - the initial scale matrix, the square root of the covariance matrix, applied to the vector of independently distributed Normal variables. This must be a lower triangular matrix with positive diagonal elements.
      gamma - a value between 0.5 and 1, that controls the speed of adaption. A lower gamma will lead to faster adoption.
      targetAcceptance - the target acceptance rate
      initialState - the initial state of the algorithm
      rnorm - the random standard Normal generator to be used
      uniform - the random long generator to be used

    • RobustAdaptiveMetropolis

      public RobustAdaptiveMetropolis(RealScalarFunction logf, double targetAcceptance, Vector initialState, RandomLongGenerator uniform)
      Constructs an instance which assumes an initial variance of 1 per variable, uses a gamma of 0.5. These parameters are suitable for many applications, though it may achieve better performance (faster execution and/or better mixing), by tweaking them.
      Parameters:
      logf - the log of the unnormalized pdf from which we wish to sample
      targetAcceptance - the target acceptance rate
      initialState - the initial state of the algorithm
      uniform - the random long generator to be used

  • Method Details

    • nextProposedState

      protected Vector nextProposedState(Vector currentState)
      Description copied from class: AbstractMetropolis
      Proposes a next state for the system.
      Specified by:
      nextProposedState in class AbstractMetropolis
      Parameters:
      currentState - the current state of the system
      Returns:
      the proposed next state

    • isProposalAccepted

      protected boolean isProposalAccepted(Vector currentState, Vector proposedState)
      Description copied from class: AbstractMetropolis
      Decides whether the given proposed state should be accepted, or whether the system should remain in it's current state.
      Specified by:
      isProposalAccepted in class AbstractMetropolis
      Parameters:
      currentState - the current state of the system
      proposedState - the proposed next state of the system
      Returns:
      whether the system should accept the proposed next state

    • S

      public Matrix S()
      Gets the tuned scaling matrix (this changes each time a new sample is drawn).
      Returns:
      the scaling matrix