mmb (version 0.13.3)

estimatePdf: Safe PDF estimation that works also for sparse random variables.

Description

Given a few observations of a random variable, this function returns an approximation of the PDF as a function. Returns also the PDF's support and argmax and works when only zero or one value was given. Depending on the used density function, two values are often enough to estimate a PDF.

Usage

estimatePdf(
  data = c(),
  densFun = function(vec) {     stats::density(vec, bw = "SJ") }
)

Arguments

data

vector of numeric data. Used to compute the empirical density of the data.

densFun

function default stats::density with bandwith 'SJ'. Function to compute the empirical density of a non-empty vector of numerical data. Note that this function needs to return the properties 'x' and 'y' as stats::density does, so that we can return the argmax.

Value

list with a function that is the empirical PDF using KDE. The list also has two properties 'min' and 'max' which represent the integratable range of that function. 'min' and 'max' are both zero if not data (an empty vector) was given. If one data point was given, then they correspond to its value -/+ .Machine$double.eps. The list further contains two numeric vectors 'x' and 'y', and a property 'argmax'. If no data was given, 'x' and 'y' are zero, and 'argmax' is NA. If one data points was given, then 'x' and 'argmax' equal it, and 'y' is set to 1. If two or more data points given, then the empirical density is estimated and 'x' and y' are filled from its estimate. 'argmax' is then set to that 'x', where 'y' becomes max.

Examples

# NOT RUN {
epdf <- mmb::estimatePdf(data = iris$Petal.Width)
print(epdf$argmax)
plot(epdf)

# Get relative likelihood of some values:
epdf$fun(0.5)
epdf$fun(1.7)
# }