[R] Rmix package and different distributions

[Ben Bolker]

nelson - <nelson1977 <at> gmail.com> writes:

> 
> hi all!
>   i want to mix a dataset that is build up from 2 distribution: an
> exponential and a normal. I can' figure out how can i use Rmix package
> to do the fitting of my data. Pheraps it si the wrong package? any
> suggestion?
> 
> thanks,
>   nelson
> 
  
  Perhaps use optim() or mle() [from the stats4 package] and
a custom-defined negative log-likelihood function ...

e.g.

nllfun <- function(p,a,mu,sigma) {
   -sum(log(p*dexp(x,a)+(1-p)*dnorm(x,mu,sigma)))
}

mle(minuslogl=nllfun,...)

[you'll need to provide starting values for the parameters]

  Please note that this is completely untested!
See Venables and Ripley's mixture-modeling example
in MASS for more details of this kind of custom fitting
(their example is a mixture of two normals, but
it should still be helpful)

  Ben Bolker