Bayesian optimization using surrogate models

bayes_opt(
  objective,
  lower,
  upper,
  type_acq = c("ei", "ucb"),
  nb_init = 10L,
  nb_iter = 90L,
  kappa = 1.96,
  method = c("standard", "direct_online", "polyak_online"),
  surrogate_model = c("rvfl", "matern52", "rvfl_emcee", "rf"),
  optim_surr = c("GCV", "loglik", "cv"),
  activation_function = c("relu", "tanh", "sigmoid"),
  type_optim = c("nlminb", "DEoptim", "msnlminb", "randsearch", "none"),
  early_stopping = FALSE,
  abs_tol = 1e-07,
  rel_tol = 0.001,
  seed = 123,
  verbose = TRUE,
  show_progress = TRUE,
  ...
)

Arguments

objective: function to be optimized
lower: lower bound for search
upper: upper bound for search
type_acq: type of acquisition function
nb_init: number of points in initial design
nb_iter: number of iterations of the algo
kappa: quantile for ucb
method: fit all, or online
surrogate_model: surrogate model
optim_surr: surrogate hyperparams fitting
activation_function: activation for bayesian rvfl
type_optim: optim for acquisition (nlminb, DEoptim, msnlminb, randsearch, none)
early_stopping: if TRUE, the algorithm will stop if the acquisition function is below a certain threshold
abs_tol: absolute tolerance for early stopping
rel_tol: relative tolerance for early stopping
seed: seed for random number generation
verbose: if TRUE, the algorithm will print progress messages
show_progress: if TRUE, the algorithm will show a progress bar
...