Package jline.api.pfqn.nc

Class Pfqn_mci

java.lang.Object

jline.api.pfqn.nc.Pfqn_mci

public final class Pfqn_mci
extends Object

Method Summary

Modifier and Type	Method	Description
static Ret.pfqnNc	pfqn_mci(Matrix D, Matrix N, Matrix Z, int I, String variant)	Monte Carlo integration for the normalizing constant.

Methods inherited from class java.lang.Object

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

Method Details

pfqn_mci

public static Ret.pfqnNc pfqn_mci(Matrix D,
 Matrix N,
 Matrix Z,
 int I,
 String variant)

Monte Carlo integration for the normalizing constant.

Variants: mci, imci (improved), amci, lhsmci and rm (repairman). amci and lhsmci use the imci tilt and differ only in how the uniforms are drawn.

amci draws ANTITHETIC pairs (u, 1-u). This does NOT reliably reduce variance here: the tilted integrand is not monotone in the exponential draws (the tilt term -(1-gamma)V decreases while the N log(VD+Z) term increases), so the pair correlation is not systematically negative; measured variance ratios against imci range from 0.54 to 1.6 across models. It is kept because it is the Ross-Wang construction, not because it is the better default. lhsmci stratifies each coordinate by Latin hypercube sampling, which IS reliably variance-reducing on the same models (ratios 0.0 to 0.48, exact quadrature in the limit of one station) at O(I log I) extra cost.

Parameters:

D - service demand matrix (M x R)

N - population vector (1 x R)

Z - think time vector (1 x R)

I - number of samples

variant - sampling variant

Returns:

the normalizing constant and its logarithm