mopt

An R package providing a collection of optimizers, designed as a companion to the rcest package. Includes a custom random-walk optimizer with adaptive step sizes, a Gibbs sampler, and wrappers for nloptr and maxLik optimizers.

Installation

devtools::install_github("bamonroe/mopt")

Features

Random Walk / MCMC Optimizer

The main mopt() function implements a random-walk optimizer with adaptive proposal standard deviations. It supports both maximum likelihood and MCMC (Bayesian) modes:

# Maximum likelihood
result <- mopt(init = c(0, 0), fn = my_loglik, method = "rwalk",
               control = list(iterations = 5000))

# MCMC with priors
result <- mopt(init = c(0, 0), fn = my_loglik, method = "mcmc",
               control = list(iterations = 10000,
                              prior_fn = my_prior,
                              prior_args = list(...)))

Gibbs Sampler

result <- gibbs_opt(init = c(a = 0, b = 0), jump = c(0.1, 0.1),
                    n = 5000, post_fn = my_posterior)

rcest Integration

When loaded alongside rcest, mopt automatically registers additional optimizers:

• nloptr: sbplx, bobyqa, cobyla, neldermead, praxis, newuoa
• maxLik: bhhh, nr
• Custom: rwalk (ML), mcmc (Bayesian)

License

MIT