hyperdirichlet: The Hyperdirichlet distribution
In hyperdirichlet: A Generalization of the Dirichlet Distribution

Description Usage Arguments Details Value Author(s) See Also Examples

Create, coerce to, or test for an object of class hyperdirichlet

1
2
3

hyperdirichlet(x, NC, pnames = character(), validated=FALSE)
is.hyperdirichlet(x)
as.hyperdirichlet(x, calculate_NC = FALSE , ...)

`x`	Object to be coerced or tested for
`NC`	Normalizing constant
`pnames`	names of the columns with length-0 default resulting in the print method using column names `p1`, `p2`, etc
`validated`	Boolean. Setting to `TRUE` is taken to mean that `x` is known to be ‘proper’ (i.e. `x` is normalizable), but the normalizing constant is not necessarily known. Default `FALSE` is taken to mean that `x` is not known to be proper: it is possible that `x` is not normalizable so cannot correspond to a PDF. Setting to `FALSE` means that the object will be passed to `is.proper()` for checking; this can be time-consuming. The flag is set to `TRUE` ab initio for `dirichlet()` and `gd()` because these distributions have an analytical expression for the normalizing constant
`calculate_NC`	Boolean, with default `FALSE` meaning not to calculate the normalizing constant and `TRUE` meaning to calculate it
`...`	Further arguments passed to `adaptIntegrate()`

To determine the normalization constant, use something like a <- as.hyperdirichlet(a,calculate_NC= TRUE).

Matrices may be coerced to a hyperdichlet object using as.hyperdirichlet(): the call is dispatched to matrix_to_HD() (qv).

If x is a matrix, be sure to specify the bernoulli argument, which is passed on to matrix_to_HD()

Functions hyperdirichlet() and as.hyperdirichlet() return a hyperdirichlet object; function is.hyperdirichlet() returns a Boolean.

Robin K. S. Hankin

B,extract,matrix_to_HD

(a <- uniform(3))
a[c(TRUE,TRUE,FALSE)] <- 0.3
## Not run: 
(a <- as.hyperdirichlet(a, calculate_NC = TRUE))  # recommended way to calculate NC

## End(Not run)

(b <- dirichlet(1:3))
as.hyperdirichlet(b)          # "forgets" the normalizing constant

## Not run: 
as.hyperdirichlet(b, TRUE)    # recalculates NC; accuracy tolerable
                              # (analytic answer = 1/60)

## End(Not run)

## Not run:   # takes a long time
op <- options()
options(warn = -1)
x <- dirichlet(rep(2,4)) + justpairs(matrix(1,4,4))
f <- function(p){p[1]>p[2]}
probability(x,f) # should be 0.5: distribution is symmetric

## End(Not run)


# following example reveals a bug in versions <= 1.3:
d <- dirichlet(1:3)
colMeans(rhyperdirichlet(1e3,d))
# result should be (1:3)/6
# implementation >= 1.4 includes a bugfix sent by Simon Byrne