dhyperdirichlet {hyperdirichlet} | R Documentation |
Probability density function for the hyperdirichlet distribution in terms of either p or e; and random sampling using Metropolis-Hastings
dhyperdirichlet_e(e, HD, include.Jacobian = TRUE) dhyperdirichlet(p, HD, include.NC = FALSE, TINY = 1e-10, log = FALSE) rhyperdirichlet(HD, n=10, start=NULL, sigma=NULL)
HD |
Object of class hyperdirichlet , or coerced thereto |
p |
Vector of length dim(HD) , notionally summing to one |
e |
Vector of length dim(HD) giving the point in e -space |
include.Jacobian |
In function dhyperdiriclet_e() ,
Boolean with default TRUE meaning to include the Jabobian of
the transform from e to p |
include.NC |
In function dhyperdirichlet_e() , Boolean with
TRUE meaning to include the normalization factor and default
FALSE meaning not to include it (it is expensive to
calculate). Note that if the normalizing factor is not known, the
function will return NA |
TINY |
In function dhyperdirichlet_p() , numeric,
specifying minimum size for elements of p via p <-
pmax(p , TINY) |
log |
In function dhyperdirichlet_p() , Boolean with
default FALSE meaning to return the probability density and
TRUE meaning to return its logarithm |
n,start,sigma |
In function rhyperdirichlet() , n is
the number of observations to take, start is the start-point
for the random walk (with default NULL meaning to use the
neutral point), and sigma is the standard deviation for the
(Gaussian) kernel, with default NULL meaning to use
1/d |
Function dhyperdirichlet()
gives the density as a function of
the p_1, ..., p_d.
Function dhyperdirichlet_e()
gives the density as a function of
the e_i. This is useful when integrating as the simplex (in
p-space) transforms to a hypercube in e-space.
Function dhyperdirichlet()
silently normalizes p
by
p <- p/sum(p)
.
The relationship between e and p is given in
e_to_p.Rd
.
Robin K. S. Hankin
dhyperdirichlet(c(1,4,3,2)/10, dirichlet(1:4)) rhyperdirichlet(dirichlet(1:3),20) diff(c(0,sort(runif(9)),1)) # random sample drawn from dirichlet(rep(1,10))