[R] Simulating a Poisson Process in R by calling C Code over .Call

[Thomas Lumley]

On Sun, 13 Jun 2010, Fabian Zäpernick wrote:

> Hi
>
> I want to write a C function for the R Code below and call it with .Call:
>
> SimPoisson <- function(lambda,tgrid,T2M)
> #Simulation eines Poissonprozesses
<snip>
> 	return(list(NT=NT,Ni=Ni,tau=tau))
> }
>
> I read the manual "writing R extensions" over and over again, but i have
> no idea, how to solve the problem with tau because i dont no the length
> of tau at the begining of the function
>

The standard approach is to start off with some reasonable guess at the length of tau and then if the vector fills up, allocate one that is twice as large and copy the current values of tau into the new vector.

In this case you can do better, since you have a very good initial guess for how long tau will be.  If you make the vector of length qpois(0.9999, lambda), then there is a 99.99% chance that it is long enough.  Using lambda+4*sqrt(lambda) is almost as good.

Yet another approach is to take advantage of the memoryless property of the Poisson process: set tau to some reasonable length, then if it fill up, return and call the function again to handle the remainder of the duration.

     -thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle