RelULSIF: Relative unconstrained least squares importance fitting

In connorbrubaker/rulsif.ts: Time Series Change Point Detection Using Relative Unconstrained Least Squares Importance Fitting (RULSIF)

Description

Estimate the ratio of probability densities that generated data in two samples (Xnu and Xde).

Usage

RelULSIF(
  Xnu,
  Xde,
  Xce = NULL,
  sigma = NULL,
  lambda = NULL,
  alpha = 0.01,
  k = 100,
  n_folds = 5
)

Arguments

Xnu	Samples from numerator probability density
Xde	Samples from denomenator probability density
Xce	Matrix of centers
sigma	Scalar or vector; Gaussian kernel bandwidth(s). Positive.
lambda	Scalar or vector; regularization parameter(s). Non-negative.
alpha	Scalar; relative parameter in [0, 1)
k	Positive integer; number of basis functions
n_folds	Integer; number of folds to use in cross fold validation

Value

List of 3

• opt_sigma: chosen sigma from CV

• opt_lambda: chosen lambda from CV

• rPE: relative Pearson divergence

Examples

X <- matrix(
    c(rnorm(50), rnorm(50, mean = 5), rnorm(50, mean = -5),
    rnorm(100), rnorm(25, mean = 3), rnorm(25, mean = -1),
    rnorm(25), rnorm(75, mean = -2), rnorm(50, mean = 4)),
    nrow = 3, ncol = 150, byrow = TRUE
)
Xnu <- X[ , 1:50]
Xde <- X[ , 51:ncol(X)]
RelULSIF(Xnu, Xde)

connorbrubaker/rulsif.ts documentation built on Dec. 19, 2021, 6:02 p.m.