ContactCenters
V. 0.9.9.

umontreal.iro.lecuyer.contactcenters.msk.stat
Class CovFMMCallCenterStat

java.lang.Object
  extended by umontreal.iro.lecuyer.contactcenters.msk.stat.CovFMMCallCenterStat

public class CovFMMCallCenterStat
extends Object

Represents a set of probes that collect covariances in matrices of functions of multiple means tallies. An instance of this class is constructed using a CallCenterStatProbes object. For each matrix of functions of multiple means tallies defined in the inner set of probes, this class can extract the covariance matrix of the functions' domain, and add these covariances into matrices of tallies. This results in averages of covariances which are useful for estimating the variance of stratified estimators.

More specifically, let $ \boldX$$ \RR^{d}_{}$ be a vector used to compute the function associated with position (r, c) in a matrix of performance measures m. Let $ \boldSigma_{\rmX}^{}$ be the matrix of covariances of $ \boldX$. We suppose that $ \boldX_{0}^{}$,…,$ \boldX_{{k-1}}^{}$ are i.i.d. and $ \boldX_{s}^{}$ is an average of ns vectors. The average covariance is

$\displaystyle \boldSigma_{\rmX}^{}$ = $\displaystyle {\frac{{1}}{{k}}}$$\displaystyle \htsum_{{s=0}}^{{k-1}}$$\displaystyle \boldSigma_{{\rmX, s}}^{}$

and the average weighted covariance is

$\displaystyle \boldSigma_{\rmX}^{}$ = $\displaystyle {\frac{{1}}{{k}}}$$\displaystyle \htsum_{{s=0}}^{{k-1}}$$\displaystyle \boldSigma_{{\rmX, s}}^{}$/ns.


Constructor Summary
CovFMMCallCenterStat(CallCenterStatProbes stat, boolean varWeighted)
          Constructs a new group of statistical probes for covariances from the inner call center statistics stat.
 
Method Summary
 void addStat()
          Adds new observations in each associated matrix of tallies.
 void covariance(PerformanceMeasureType pm, int row, int col, DoubleMatrix2D cov)
          Returns the covariance matrix for the function of multiple means tally corresponding to the element (row, col) of the matrix of performance measures pm.
 void init()
          Initializes every matrix of tallies encapsulated in this object.
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

CovFMMCallCenterStat

public CovFMMCallCenterStat(CallCenterStatProbes stat,
                            boolean varWeighted)
Constructs a new group of statistical probes for covariances from the inner call center statistics stat. If varProp is true, the returned covariances correspond to the proportional allocation in stratification. Otherwise, they depend on the number of observations in each stratum.

Parameters:
stat - the call center statistical object.
varWeighted - the proportional allocation indicator.
Method Detail

covariance

public void covariance(PerformanceMeasureType pm,
                       int row,
                       int col,
                       DoubleMatrix2D cov)
Returns the covariance matrix for the function of multiple means tally corresponding to the element (row, col) of the matrix of performance measures pm. If covariances are not queried for proportional allocation, the covariances are divided by the number of observations in the encapsulated tallies.

Parameters:
pm - the type of performance measure.
row - the row in the matrix.
col - the column in the matrix.
cov - the 2D matrix filled with covariances.

init

public void init()
Initializes every matrix of tallies encapsulated in this object.


addStat

public void addStat()
Adds new observations in each associated matrix of tallies. This method is called at the end of each stratum or macroreplication and extracts the covariances from the matrices of functions of multiple means tallies. It then adds the covariances to the encapsulated matrices of tallies. If covariances are not queried for proportionnal allocation, the covariances are divided by the number of observations before they are added to the tallies, resulting in a weighted sum of covariances.


ContactCenters
V. 0.9.9.

To submit a bug or ask questions, send an e-mail to Richard Simard.