private void initialize(IDensityKernel kernel, double[][] observations,
double[] weights, int[] repeats, double[,] smoothing)
{
if (smoothing == null)
{
smoothing = SilvermanRule(observations, weights, repeats);
}
if (kernel == null)
{
kernel = new GaussianKernel(Dimension);
}
this.kernel = kernel;
this.samples = observations;
this.smoothing = smoothing;
this.determinant = smoothing.Determinant();
this.weights = weights;
this.repeats = repeats;
this.chol = new CholeskyDecomposition(smoothing);
if (weights != null)
{
this.type = WeightType.Fraction;
this.numberOfSamples = samples.Length;
this.sumOfWeights = weights.Sum();
}
else if (repeats != null)
{
this.type = WeightType.Repetition;
this.numberOfSamples = repeats.Sum();
this.sumOfWeights = 1.0;
}
else
{
this.type = WeightType.None;
this.numberOfSamples = samples.Length;
this.sumOfWeights = 1.0;
}
this.mean = null;
this.variance = null;
this.covariance = null;
}
private void initialize(IDensityKernel kernel, double[][] observations, double[,] smoothing)
{
if (smoothing == null)
{
smoothing = SilvermanRule(observations);
}
if (kernel == null)
{
kernel = new GaussianKernel(Dimension);
}
this.kernel = kernel;
this.samples = observations;
this.smoothing = smoothing;
this.determinant = smoothing.Determinant();
this.mean = null;
this.variance = null;
this.covariance = null;
}
private void initialize(IDensityKernel kernel, double[][] observations, double[,] smoothing)
{
if (smoothing == null)
{
// Silverman's rule
// - http://en.wikipedia.org/wiki/Multivariate_kernel_density_estimation
double[] sigma = Statistics.Tools.StandardDeviation(observations);
double d = sigma.Length;
double n = observations.Length;
smoothing = new double[sigma.Length, sigma.Length];
for (int i = 0; i < sigma.Length; i++)
{
double a = Math.Pow(4 / (d + 2), 1 / (d + 4));
double b = Math.Pow(n, -1 / (d + 4));
smoothing[i, i] = a * b * sigma[i];
}
}
if (kernel == null)
{
kernel = new GaussianKernel(Dimension);
}
this.kernel = kernel;
this.samples = observations;
this.smoothing = smoothing;
this.determinant = smoothing.Determinant();
this.mean = null;
this.variance = null;
this.covariance = null;
}
/// <summary>
/// Creates a new Empirical Distribution from the data samples.
/// </summary>
///
/// <param name="kernel">The kernel density function to use.
/// Default is to use the <see cref="GaussianKernel"/>.</param>
/// <param name="samples">The data samples.</param>
/// <param name="weights">The fractional weights to use for the samples.
/// The weights must sum up to one.</param>
/// <param name="smoothing">
/// The kernel smoothing or bandwidth to be used in density estimation.
/// By default, the normal distribution approximation will be used.</param>
///
public MultivariateEmpiricalDistribution(IDensityKernel kernel, double[][] samples, double[] weights, double[,] smoothing)
: base(samples[0].Length)
{
this.initialize(kernel, samples, weights, null, smoothing);
}
/// <summary>
/// Creates a new Empirical Distribution from the data samples.
/// </summary>
///
/// <param name="kernel">The kernel density function to use.
/// Default is to use the <see cref="GaussianKernel"/>.</param>
/// <param name="samples">The data samples forming the distribution.</param>
///
public MultivariateEmpiricalDistribution(IDensityKernel kernel, double[][] samples)
: base(samples[0].Length)
{
this.initialize(kernel, samples, null, null, null);
}
private void initialize(IDensityKernel kernel, double[][] observations, double[,] smoothing)
{
if (smoothing == null)
{
// Silverman's rule
// - http://en.wikipedia.org/wiki/Multivariate_kernel_density_estimation
double[] sigma = Statistics.Tools.StandardDeviation(observations);
double d = sigma.Length;
double n = observations.Length;
smoothing = new double[sigma.Length, sigma.Length];
for (int i = 0; i < sigma.Length; i++)
{
double a = Math.Pow(4 / (d + 2), 1 / (d + 4));
double b = Math.Pow(n, -1 / (d + 4));
smoothing[i, i] = a * b * sigma[i];
}
}
if (kernel == null)
kernel = new GaussianKernel(Dimension);
this.kernel = kernel;
this.samples = observations;
this.smoothing = smoothing;
this.determinant = smoothing.Determinant();
this.mean = null;
this.variance = null;
this.covariance = null;
}
/// <summary>
/// Creates a new Empirical Distribution from the data samples.
/// </summary>
///
/// <param name="kernel">The kernel density function to use.
/// Default is to use the <see cref="GaussianKernel"/>.</param>
/// <param name="samples">The data samples.</param>
/// <param name="smoothing">
/// The kernel smoothing or bandwidth to be used in density estimation.
/// By default, the normal distribution approximation will be used.</param>
///
public MultivariateEmpiricalDistribution(IDensityKernel kernel, double[][] samples, double[,] smoothing)
: base(samples[0].Length)
{
this.initialize(kernel, samples, smoothing);
}