R Mixture Distribution Function

Here’s a useful little utility: it is an R mixture distribution function. I needed a basic version of this for investigating some of Taleb’s ideas, and decided to put together an easy-to-use general version.

R Mixture Distribution Function: Code

[code language=”r” title=”Mixture Distribution”]# Mixture distribution

rMixDist <- function(n=1000, DISTs=c(rnorm, rnorm),
pars=list(c("mean"=0, "sd"=1), c("mean"=0, "sd"=2)),
probs=c(0.99, 0.01)){
# Ian Rayner 2016-06
if (sum(probs) != 1) stop("Probabilities should sum to 1")
if (any(sapply(list(DISTs, pars), length) – length(probs) != 0))
stop("Length of DISTs, pars and probs must match")
counts <- table(factor(sample(1:length(probs), n, T, prob=probs),
levels=1:length(probs)))
popSamp <- c()
for (i in 1:length(counts)){
argList <- as.list(c("n"=unname(counts[i]), pars[[i]]))
popSamp <- c(popSamp, do.call(DISTs[[i]], argList))
}
return(popSamp[sample(n, n)])
}[/code]

Using the typical R notation, the function generates a random sample according to the mixture parameters passed to the function.

If the user enters a set of probabilities that do not add up to one, or a set of distributions, parameters, and probabilities that are not the same in number, the function throws an error.

R Mixture Distribution Function: Usage Example

If we want to generate a sample with 100,000 elements that is a mixture of two normal distributions (~N(1, 0.1) and ~N(2, 0.2)) and a log normal distribution ~LogN(0, 0.3) in a ratio 80:17:3, we would call the following function:

rMixDist(100000, DISTs=c(rnorm, rnorm, rlnorm), pars=list(c(1, 0.1), c(2, 0.2), c(0, 0.3)), probs=c(0.8, 0.17, 0.03))

This results in the following sample:

Enjoy!