org.knowceans.util
Class ArmSampler

java.lang.Object
  org.knowceans.util.ArmSampler

public abstract class ArmSampler
extends java.lang.Object

ArmSampler implements an adaptive rejection Metropolis sampler (ARMS) that can sample from virtually any univariate distribution. The method performs best if a log-concave density is being sampled from, but it also works for other densities, for which an additional Metropolis step is inserted. The greater the difference to log-concave shape, the more Metropolis rejections must be expected.

This implementation is a port of the original c / fortran implementation by Wally Gilks available at http://www.mrc-bsu.cam.ac.uk/BSUsite/Research/ars.shtml.

Please acknowledge this work by referencing the relevant scientific literature and program code (Web: http://www.arbylon.net/projects).

References:

Gilks, W. R. (1992) Derivative-free adaptive rejection sampling for Gibbs sampling. Bayesian Statistics 4, (eds. Bernardo, J., Berger, J., Dawid, A. P., and Smith, A. F. M.) Oxford University Press.

Gilks, W. R., Best, N. G. and Tan, K. K. C. (1995) Adaptive rejection Metropolis sampling. Applied Statistics, 44, 455-472.

Gilks, W. R. and Wild, P. (1992) Adaptive rejection sampling for Gibbs sampling. Applied Statistics 41, pp 337-348.

Author:

gregor heinrich

Field Summary

static int	DEREF dereference for ported c pointers
static double	EYEPS critical relative exp(y) difference
static double	XEPS critical relative x-value difference
static double	YCEIL maximum y avoiding overflow in exp(y)
static double	YEPS critical y-value difference

Constructor Summary

ArmSampler()
init (nothing to do)

Method Summary

double[]	arms(java.lang.Object params, double[] xinit, int ninit, double[] xl, double[] xr, double[] convex, int npoint, boolean dometrop, double[] xprev, double[] xsamp, int nsamp, double[] qcent, double[] xcent, int ncent, int[] neval) to perform derivative-free adaptive rejection sampling with metropolis step
double	armsSimple(java.lang.Object params, int ninit, double[] xl, double[] xr, boolean dometrop, double[] xprev) adaptive rejection metropolis sampling - simplified argument list
abstract double	logpdf(double x, java.lang.Object params) Abstract function to implement the log pdf.

Methods inherited from class java.lang.Object

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

Field Detail

DEREF

public static final int DEREF

dereference for ported c pointers

See Also:

Constant Field Values

XEPS

public static final double XEPS

critical relative x-value difference

See Also:

Constant Field Values

YEPS

public static final double YEPS

critical y-value difference

See Also:

Constant Field Values

EYEPS

public static final double EYEPS

critical relative exp(y) difference

See Also:

Constant Field Values

YCEIL

public static final double YCEIL

maximum y avoiding overflow in exp(y)

See Also:

Constant Field Values

Constructor Detail

ArmSampler

public ArmSampler()

init (nothing to do)

Method Detail

logpdf

public abstract double logpdf(double x,
                              java.lang.Object params)

Abstract function to implement the log pdf.

Parameters:

x -

params -

Returns:

armsSimple

public double armsSimple(java.lang.Object params,
                         int ninit,
                         double[] xl,
                         double[] xr,
                         boolean dometrop,
                         double[] xprev)
                  throws java.lang.Exception

adaptive rejection metropolis sampling - simplified argument list

Parameters:

params[] - parameters of the pdf to be sampled

ninit - : number of starting values to be used

xl[] - : left bound ([] for pointer)

xr[] - : right bound ([] for pointer)

dometrop - : whether metropolis step is required

xprev[] - : current value from markov chain ([] for pointer)

Returns:

: sampled value

Throws:

java.lang.Exception

arms

public double[] arms(java.lang.Object params,
                     double[] xinit,
                     int ninit,
                     double[] xl,
                     double[] xr,
                     double[] convex,
                     int npoint,
                     boolean dometrop,
                     double[] xprev,
                     double[] xsamp,
                     int nsamp,
                     double[] qcent,
                     double[] xcent,
                     int ncent,
                     int[] neval)
              throws java.lang.Exception

to perform derivative-free adaptive rejection sampling with metropolis step

Parameters:

params - parameters of the pdf to be sampled

xinit[] - : starting values for x in ascending order

ninit - : number of starting values supplied

xl[] - : left bound ([] for pointer)

xr[] - : right bound ([] for pointer)

convex[] - : adjustment for convexity ([] for pointer)

npoint - : maximum number of envelope points

dometrop - : whether metropolis step is required

xprev[] - : previous value from markov chain ([] for pointer)

xsamp[] - : to store sampled values

nsamp - : number of sampled values to be obtained

qcent[] - : percentages for envelope centiles

xcent[] - : to store requested centiles

ncent - : number of centiles requested

neval[] - : on exit, the number of function evaluations performed ([] for pointer)

Returns:

xsamp[] : sampled values

Throws:

java.lang.Exception