miesmuschel (version 0.0.4-3)

mies_step_fidelity: Re-Evaluate Existing Configurations with Higher Fidelity

Description

As part of the "rolling-tide" multifidelity-setup, do reevaluation of configurations with higher fidelity that have survived lower-fidelity selection. The evaluations are done as part of the current generation, so the dob value is not increased.

This function should only be called when doing rolling-tide multifidelity, and should not be part of the MIES cycle otherwise.

Usage

mies_step_fidelity(
  inst,
  budget_id,
  fidelity,
  current_gen_only = FALSE,
  fidelity_monotonic = TRUE,
  additional_components = NULL
)

Value

invisible

data.table: the performance values returned when evaluating the offspring values through eval_batch.

Arguments

inst

(OptimInstance)
Optimization instance to evaluate.

budget_id

(character(1))
Budget component that is set to the fidelity value.

fidelity

(atomic(1))
Atomic scalar indicating the value to be assigned to the budget_id component of offspring.

current_gen_only

(logical(1))
Whether to only re-evaluate survivors individuals generated in the latest generation (TRUE), or re-evaluate all currently alive individuals (FALSE). In any case, only individuals that were not already evaluated with the chosen fidelity are evaluated, so this will usually only have an effect when the fidelity of surviving individuals changed between generations.

fidelity_monotonic

(logical(1))
Whether to only re-evaluate configurations for which the fidelity would increase. Default TRUE.

additional_components

(ParamSet | NULL)
Additional components to optimize over, not included in search_space, but possibly used for self-adaption. This must be the ParamSet of mies_init_population()'s additional_component_sampler argument.

See Also

Other mies building blocks: mies_evaluate_offspring(), mies_generate_offspring(), mies_get_fitnesses(), mies_init_population(), mies_select_from_archive(), mies_survival_comma(), mies_survival_plus()

Examples

library("bbotk")
lgr::threshold("warn")

# Define the objective to optimize
objective <- ObjectiveRFun$new(
  fun = function(xs) {
    z <- exp(-xs$x^2 - xs$y^2) + 2 * exp(-(2 - xs$x)^2 - (2 - xs$y)^2)
    list(Obj = z)
  },
  domain = ps(x = p_dbl(-2, 4), y = p_dbl(-2, 4)),
  codomain = ps(Obj = p_dbl(tags = "maximize"))
)

# Get a new OptimInstance
oi <- OptimInstanceSingleCrit$new(objective,
  terminator = trm("evals", n_evals = 100)
)

budget_id = "y"

# Create an initial population with fidelity ("y") value 1
mies_init_population(oi, mu = 2, budget_id = budget_id, fidelity = 1)

oi$archive

# Re-evaluate these individuals with higher fidelity
mies_step_fidelity(oi, budget_id = budget_id, fidelity = 2)

oi$archive

# The following creates a new generation without killing the initial
# generation
offspring = data.frame(x = 0:1)
mies_evaluate_offspring(oi, offspring = offspring, budget_id = budget_id,
  fidelity = 3)

oi$archive

# Re-evaluate only individuals from last generation by setting current_gen_only
mies_step_fidelity(oi, budget_id = budget_id, fidelity = 4,
  current_gen_only = TRUE)

oi$archive

# Default: Re-evaluate all that *increase* fidelity: Only the initial
# population is re-evaluated here.
mies_step_fidelity(oi, budget_id = budget_id, fidelity = 3)

oi$archive

# To also re-evaluate individuals with *higher* fidelity, use
# 'fidelity_monotonic = FALSE'. This does not re-evaluate the points that already have
# the requested fidelity, however.
mies_step_fidelity(oi, budget_id = budget_id, fidelity = 3, fidelity_monotonic = FALSE)

oi$archive