G'day Jose,
On Fri, 29 Oct 2010 11:25:05 +0200
Jos? Manuel Gavil?n Ruiz <gavi at us.es> wrote:
> Hello, I want to maximize a likelihood function expressed as an
> integral that can not be symbolically evaluated. I expose my problem
> in a reduced form.
>
> g<- function(x){
> integrand<-function(y) {exp(-x^2)*y}
> g<-integrate(integrand,0,1)
> }
> h<-function(x) log((g(x)))
>
> g is an object of the class function, but g(2) is a integrate object,
> I can print(g(2)) an get a result, but
> if I try h(2) R says is a nonnumeric argument for a mathematical
> function.
R> print(g(2))
0.00915782 with absolute error < 1.0e-16
Indeed print(g(2)) gives an output, but it is obviously not just a
numeric number but something formatted, presumably based on the class
of the returned object from g(2) and the values that this object
contains. (You may want to read up on R's way(s) to object oriented
programming).
So what does g() return?
R> str(g(2))
List of 5
$ value : num 0.00916
$ abs.error : num 1.02e-16
$ subdivisions: int 1
$ message : chr "OK"
$ call : language integrate(f = integrand, lower = 0, upper = 1)
- attr(*, "class")= chr "integrate"
The object is a list with several values, and class "integrate".
> My goal is to maximize h. what?s wrong?
I guess you want to pass only the component value to h().
R> g <- function(x){
+ integrand <- function(y) {exp(-x^2)*y}
+ integrate(integrand, 0, 1)$value}
R> g(2)
[1] 0.00915782
R> h(g(2))
[1] -0.693231
HTH,
Cheers,
Berwin