opt_lin_bw {cylcop}

Find the Optimal Bandwidth for a Linear Kernel Density Estimate

Description

This function wraps stats::bw.ucv() and stats::bw.nrd() of the 'stats' package, simplifying their inputs. For more control, these 'stats' functions could be used directly.

Usage

opt_lin_bw(x, method = c("cv", "nrd"))

Arguments

x

numeric vector of linear measurements.

method

character string describing the method used to find the optimal bandwidth. Either "cv" (cross-validation), or "nrd" (rule-of-thumb estimate).

Details

The normal reference distribution (nrd) method involves matching a normal distribution to the data using an empirical measure of spread. The optimal bandwidth for that normal distribution can then be exactly calculated by minimizing the mean integrated square error. method="cv" finds the optimal bandwidth using unbiased cross-validation.

Value

A numeric value, the optimized bandwidth.

See Also

stats::bw.ucv(), stats::bw.nrd() opt_lin_bw().

Examples

require(graphics)
set.seed(123)
n <- 1000

x <- rweibull(n, shape = 10)
bw1 <- opt_lin_bw(x = x, method="nrd")
bw2 <- opt_lin_bw(x = x, method="cv")

dens1 <- fit_steplength(x = x, parametric = FALSE, bandwidth = bw1)
dens2 <- fit_steplength(x = x, parametric = FALSE, bandwidth = bw2)
true_dens <- dweibull(seq(0,max(x),length.out = 200), shape = 10)

plot(seq(0,max(x),length.out = 200), true_dens, type = "l")
lines(dens1$x, dens1$y, col = "red")
lines(dens2$x, dens2$y, col = "green")