[R] Bayesian question (problem using adapt)

[Tamas K Papp]

On Sat, Nov 11, 2006 at 11:19:46AM -0500, cmdicaso at ncsu.edu wrote:

>  kernel1 <- function(theta, param) { exp(loglik(theta, param)) + log(10000)}
> 
>  kernel2 <- function(theta, param) {( exp(loglik(theta, param)) +
> log(10000)) * gm(theta, param)  }
> 
>  kernel2(theta0, param0)
> 
> 
> 
>  adapt(4, lower = c(0, 0, 0, 0), upper = c(1, 1, 1, 1), functn =
> kernel2(theta0, param0))

I could not walk through your code because you failed to create a
simple reproducible example (eg the data is missing), but
kernel2(theta0,param0) is no longer a function --- kernel2 is, you
need to just supply that.

A couple of remarks:

- instead of "function undefined", you might just want to return NA or
NaN as is standard in R.

- why do you multiply by 10000?

- are you just integrating the posterior?  why?

- have you considered MCMC?  On state spaces bigger than 2 dimension,
  it is usually preferable to numerical integration (but since I don't
  know your model in detail, there might be something special about
  it).

Best,

Tamas