Package 'edfun'

Creating Empirical Distribution Functions

Description

A function for creating a set of (one dimensional) empirical distribution functions (density, CDF, inv-CDF, and random number generator). This is either based on a vector of observations from the distribution, or a density function.

Usage

edfun(x, support = range(x), dfun, qfun_method = NULL, ...)

Arguments

x	numeric vector of data or (in case density is not NULL) a sequance of values for which to evaluate the density function for creating the inv-CDF. Also, the rfun will be based on the inverse CDF on uniform distribution (inv-CDF(U[0,1]) - which is "better" than using sample, if we have the density).
support	a 2d numeric vector giving the boundaries of the distribution. Default is the range of x. This is used in qfun to decide how to work with extreme cases of q->0|1.
dfun	a density function. If supplied, this creates a different pfun (which now relies on integrate) and rfun (which will now rely on inv-CDF(U[0,1])). If missing, then it is created using density. If NULL then it is not created.
qfun_method	can get a quantile function to use (for example "quantile"), with the first parameter accepts the data (x) and the second accepts probs (numeric vector of probabilities with values in [0,1]). If it is NULL (the default) then the quantiles are estimated using approxfun from predicting the x values from the pfun(x) values.
...	ignored

Value

A list with 4+ components: dfun, pfun, qfun and rfun. The 5th componont is pfun_integrate_dfun which is NUNLL if dfun is not supplied. If it is supplied, it returns a function that relies on integrate of dfun for returning pfun. Since this method is VERY slow, it is not returned within pfun. Instead, pfun will pre-compute pfun_integrate_dfun on all values of x.

Each component is a function to perform the usual tasks of distributions.

Examples

set.seed(2016-08-18)
x <- rnorm(100)
x_funs <- edfun(x)
x_funs$qfun(0) # -2.6

# for extreme cases, we can add the support vector
x_funs <- edfun(x, support = c(-Inf, Inf))
x_funs$qfun(0) # -Inf

f <- x_funs$dfun
curve(f, -2,2)

f <- x_funs$pfun
curve(f, -2,2)

f <- x_funs$qfun
curve(f, 0,1)

f <- x_funs$rfun
hist(f(1000))

---