In USCbiostats/amcmc: A friendly MCMC framework

To generate kernels efficiently, we can make use of the following:

1. Avoid creating/destructing vectors: It is wiser to pass an already built vector, so handles pointers and returns void, instead of actually creating a vector within the function and assigning that to an outcome, and later destroying the value.

2. Potentially, computing the acceptance within the kernel function, which implies that we need to pass the loglikelihood function as well. This could be better in terms of reading the code.

3. The functions should receive an object of class NumericVector and return double, no other arguments. This can be done easily by creating wrappers before passing the function when calling C++. A problem arises if we want to call the amcmc from within C++ and we need to pass more arguments.

4. Because of the previous it makes sense to make the function more modular, or more over, more flexible. Instead of working with vectors, a function can be written such that it receives a list of arguments instead in which the first is assumed to be the parameters over which we are working on, for example:

   ```r

   The actual function

   fun <- function(arg0, arg1) { dnorm(arg0, mean=arg1) }

   The function that we call

   myfun <- function(args) { do.call(fun, args) }

   How the kernel can look

   kern <- function(args) { update(args[[1]]) } ```

USCbiostats/amcmc documentation built on Sept. 22, 2023, 5:07 a.m.