public static IArtaProcess CreateArtaProcess(RealDistribution distribution, double[] artaCorrelationCoefficients, RandomGenerator random) //throws NonFeasibleCorrelationException, NotStationaryException
{
AbstractArtaProcess arta = null;
if (artaCorrelationCoefficients == null || artaCorrelationCoefficients.Length == 0)
{
double[] noCorrelation = { 0.0 };
artaCorrelationCoefficients = noCorrelation;
}
// check feasibility
FeasibilityTest ft = new FeasibilityTest(distribution);
ft.CheckFeasibility(artaCorrelationCoefficients);
// check if correlation matrix is positive definite, if not CholeskyDecomposition throws Error
new CholeskyDecomposition(AutoCorrelation.GetCorrelationMatrix(artaCorrelationCoefficients));
if (distribution is NormalDistribution)
{
arta = CreateArtaProcessN((NormalDistribution)distribution, artaCorrelationCoefficients, random);
}
else if (distribution is UniformRealDistribution)
{
arta = createArtaProcessU((UniformRealDistribution)distribution, artaCorrelationCoefficients, random);
}
else
{
arta = CreateArtaProcessG(distribution, artaCorrelationCoefficients, random);
}
return(arta);
}
private static ArtaProcessGeneral CreateArtaProcessG(RealDistribution distribution, double[] artaCorrelationCoefficients, RandomGenerator random) //throws NotStationaryException
{
ArtaProcessGeneral arta = null;
AutocorrelationFitter fitter = new AutocorrelationFitter(distribution);
double[] arCorrelationCoefficients = fitter.FitArAutocorrelations(artaCorrelationCoefficients, DEFAULT_ERROR);
ArProcess ar = ArProcessFactory.CreateArProcess(arCorrelationCoefficients, random);
arta = new ArtaProcessGeneral(ar, distribution);
return(arta);
}
private double CalculateMinimumFeasibleBivariateCorrelation(RealDistribution distribution)
{
double minFBC = 0.0;
// consider special cases, final else is general approach
if (distribution is NormalDistribution || distribution is UniformRealDistribution || distribution is TDistribution)
{
minFBC = -1.0;
}
else if (distribution is ExponentialDistribution)
{
minFBC = 1.0 - Math.PI * Math.PI / 6.0;
}
else
{
double mean = distribution.getNumericalMean();
double variance = distribution.getNumericalVariance();
double integral = integrator.integrate(IterativeLegendreGaussIntegrator.DEFAULT_MAX_ITERATIONS_COUNT, new UnivariatDistributionFunction(distribution), 0.0, 1.0);
minFBC = (integral - mean * mean) / variance;
}
return(minFBC);
}
private UnivariatDistributionFunction(RealDistribution distribution)
{
this.distribution = distribution;
}
public FeasibilityTest(RealDistribution distribution) : this(new IterativeLegendreGaussIntegrator(5, IterativeLegendreGaussIntegrator.DEFAULT_RELATIVE_ACCURACY, IterativeLegendreGaussIntegrator.DEFAULT_ABSOLUTE_ACCURACY), distribution)
{
}
public FeasibilityTest(UnivariateIntegrator integrator, RealDistribution distribution)
{
this.integrator = integrator;
this.minimumFeasibleBivariateCorrelation = CalculateMinimumFeasibleBivariateCorrelation(distribution);
}
public ArtaCorrelationEstimator(RealDistribution dist)
{
this.distribution = dist;
}
public double nextInversionDeviate(RealDistribution distribution)
{
return(distribution.inverseCumulativeProbability(nextUniform(0, 1)));
}
public static IArtaProcess CreateArtaProcess(RealDistribution distribution, double[] artaCorrelationCoefficients)// throws NonFeasibleCorrelationException, NotStationaryException
{
return(CreateArtaProcess(distribution, artaCorrelationCoefficients, new RandomAdaptor(new MersenneTwister())));
}
public ArtaProcessGeneral(ArProcess ar, RealDistribution distribution) : base(ar)
{
this.distribution = distribution;
// TODO Auto-generated constructor stub
}
public AutocorrelationFitter(RealDistribution dist)
{
this.artaCorrelationEstimator = new ArtaCorrelationEstimator(dist);
}