Use bbmle with vector arguments: mle2 (already run with optim)

bbmle:mle2I'm having some trouble using the function when trying to do a regression . To illustrate my problem, I came up with a toy example.

We define negative log-likelihood for a Poisson distribution (or any custom distribution):

LL <- function(beta, z, x){
  -sum(stats::dpois(x, lambda = exp(z %*% beta), log = TRUE))
}

In the code above, betais the parameter vector I want to estimate the zmodel/design matrix, xor the variable I'm interested in.

Then I generate some random data to use:

set.seed(2)
age <- round(exp(rnorm(5000, mean = 2.37, sd = 0.78) - 1))
claim <- rpois(5000, lambda = 0.07

optimI can do regression analysis easily . This is the interceptonly model:

z1 <- model.matrix(claim ~ 1)
optim(par = 0, fn = LL, z = z1, x = claim)

Here is the intercept + agemodel:

z2 <- model.matrix(claim ~ age)
optim(par = c(0, 0), fn = LL, z = z2, x = claim)

The way to evaluate a large number of different models is as simple as specifying the model matrix. How can I make it work with the mle2functions in the package bbmle?

I can do, if beta1D:

mle2(minuslogl = function(beta){ LL(beta = beta, z = z1, x = claim) },
  start = list(beta = 0))

But if it beta's a vector, then I have a problem:

mle2(
  minuslogl = function(beta){ LL(beta = beta, z = z2, x = claim) },
  start = list(beta = c(0, 0)),
  vecpar = T,
  parnames = colnames(z2)
  )

I can't get the syntax right and I can't find any examples in the documentation or vignette to help me. The problem must be that it betais now a vector. The documentation recommends using a vecpar = Tparameter that is "and optim" compatible . Any hints would be greatly appreciated.

Also, is there a way to pass the zsum parameter in a more elegant way like I do in the log likelihood function ?xmle2optim