adeba: Make ADEBA density estimate

In adeba: Adaptive Density Estimation by Bayesian Averaging

Description

Make ADEBA density estimate

Usage

adeba(data, adaptive = TRUE, beta = 0.5, parallel, na.rm = FALSE, ...)

Arguments

data	Dataset with examples as rows and dimensions as columns, supplied as a matrix or something that can be converted to a matrix, like a vector, data.frame or data.table.
adaptive	Whether to used fixed bandwidths identical for all kernels, or adaptive bandwidths unique to each kernel. Can also be specified as a positive integer to iterate beyond the adaptive estimate (see the original publication).
beta	The level of adaptiveness. The default value of 0.5 corresponds to Silverman's square root law, which is suitable for the normal distribution. Can also be specified as multiple equispaced values, automatically weighted by the Bayesian machinery (the equispace requirement is for the integral approximation to hold).
parallel	Whether to use multiple CPU cores for calculation, using mclapply and mcMap of the parallel package. To control the number of cores please use the mc.cores option. If unset, all cores will be used.
na.rm	Whether to let missing values break execution (FALSE) or to remove them from calculations (TRUE).
...	Sent to the internal function make.adeba.

Value

A density estimate.

Author(s)

Christofer B<c3><a4>cklin

See Also

predict.adeba, render, radeba

Examples

# Univariate
x.train <- c(rnorm(20, 0, .5), rnorm(20, 2, 2))
f <- adeba(x.train, beta = 0:2/2)

curve((dnorm(x, 0, .5) + dnorm(x, 2, 2))/2, -2, 7, lwd=5, col="grey85")
plot(f, type="both", add=TRUE)
points(f)

# Bi-variate
x.train <- 3 + sweep(matrix(rnorm(60), 30), 2, 1:2, "*") %*% matrix(c(1, .4, .4, 1), 2)
x.test <- 3 + sweep(matrix(rnorm(40), 20), 2, 1:2, "*") %*% matrix(c(1, .4, .4, 1), 2)
f <- adeba(x.train, adaptive=FALSE)
f <- render(f)

plot(f, type="both")
points(x.test[,1], x.test[,2], cex=10*predict(f, x.test), pch=19)
legend("topleft", c("Traning", "Test"), pch=c(1,19), bg="white")

# Draw random sample from the estimated density
x.new <- radeba(400, f)
plot(f)
points(x.new[,1], x.new[,2])

# Slice the distribution
plot(0, 0, type="n", xlim=c(-4, 12), ylim=c(0, 0.15))
for(i in -6:16){
    f <- render(f, list(seq(-4, 12, length.out=200), i))
    plot(f, col=hsv(h=(i+6)/30), add=TRUE)
}
plot(f, type="data")

# See package `adebaExtra` for how to plot in full 3d

adeba documentation built on May 2, 2019, 3:21 a.m.