mlr3mbo

A new R6 and much more modular implementation for single- and multicrit Bayesian optimization. We are really not done here! Very prelim code, so beware!

List of some Ideas

Simple Example

Currently broken

set.seed(1)
library(bbotk)
library(mlr3)
library(mlr3mbo)
library(paradox)
library(mlr3learners)

obfun = ObjectiveRFun$new(
  fun = function(xs) sum(unlist(xs)^2),
  domain = ParamSet$new(list(ParamDbl$new("x", -5, 5))),
  id = "test"
)

terminator = trm("evals", n_evals = 20)

instance = OptimInstanceSingleCrit$new(
  objective = obfun,
  terminator = terminator
)

design = generate_design_lhs(obfun$domain, 4)$data
instance$eval_batch(design)

surrogate = SurrogateSingleCritLearner$new(learner = lrn("regr.km"))
acqfun = AcqFunctionCB$new(surrogate = surrogate)
acqopt = AcqOptimizerFromOptimizer$new(
  opt("random_search", batch_size = 100),
  trm("evals", n_evals = 100)
)
bayesop_soo(instance, acqfun, acqopt)
plot(y~batch_nr, instance$archive$data[batch_nr>1,], type = "b")

Multipoint through constant liar

Currently broken example:

instance = OptimInstanceSingleCrit$new(
  objective = obfun,
  terminator = terminator
)
design = generate_design_lhs(obfun$domain, 4)$data
instance$eval_batch(design)

surrogate = SurrogateSingleCritLearner$new(learner = lrn("regr.km"))
acqfun = AcqFunctionCB$new(surrogate = surrogate)
acqopt = AcqOptimizerFromOptimizer$new(
  opt("random_search", batch_size = 100),
  trm("evals", n_evals = 100)
)
bayesop_mpcl(instance, acqfun, acqopt)
plot(y~batch_nr, instance$archive$data()[batch_nr>1,], type = "b")