R/mies_methods.R
mies_survival_comma.RdChoose survivors during a MIES iteration using the "Comma" survival strategy, i.e.
selecting survivors from the latest generation only, using a Selector operator, and choosing
"elites" from survivors from previous generations using a different Selector operator.
When n_elite is greater than the number of alive individuals from previous generations,
then all these individuals from previous generations survive. In this case, it is
possible that more than mu - n_elite individuals from the current generation survive.
Similarly, when mu is greater
than the number of alive individuals from the last generation, then all these individuals survive.
mies_survival_comma(inst, mu, survival_selector, n_elite, elite_selector, ...)(OptimInstance)
Optimization instance to evaluate.
(integer(1))
Population target size, non-negative integer.
(Selector)Selector operator that selects surviving individuals depending on configuration values
and objective results. When survival_selector$operate() is called, then objectives that
are being minimized are multiplied with -1 (through mies_get_fitnesses), since Selectors always try to maximize fitness.
The Selector must be primed on inst$search_space; this includes the "budget" component
when performing multi-fidelity optimization.
The given Selector may not return duplicates.
(integer(1))
Number of individuals to carry over from previous generations. n_elite individuals will be selected
by elite_selector, while mu - n_elite will be selected by survival_selector from the most
recent generation. n_elite may be 0 (no elitism), in which case only individuals from the newest
generation survive. n_elite must be strictly smaller than mu to permit any optimization progress.
(Selector)Selector operator that selects "elites", i.e. surviving individuals from previous generations,
depending on configuration values
and objective results. When elite_selector$operate() is called, then objectives that
are being minimized are multiplied with -1 (through mies_get_fitnesses()), since Selectors always try to maximize fitness.
The Selector must be primed on inst$search_space; this includes the "budget" component
when performing multi-fidelity optimization.
The given Selector may not return duplicates.
(any)
Ignored, for compatibility with other mies_survival_* functions.
invisible data.table: The value of inst$archive$data, changed
in-place with eol set to the current generation for non-survivors.
Other mies building blocks:
mies_evaluate_offspring(),
mies_generate_offspring(),
mies_get_fitnesses(),
mies_init_population(),
mies_select_from_archive(),
mies_step_fidelity(),
mies_survival_plus()
set.seed(1)
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)
)
mies_init_population(inst = oi, mu = 3)
# Usually the offspring is generated using mies_generate_offspring()
# Here shorter for demonstration purposes.
offspring = generate_design_random(oi$search_space, 3)$data
mies_evaluate_offspring(oi, offspring = offspring)
# State before: different generations of individuals. Alive individuals have
# 'eol' set to 'NA'.
oi$archive
#> <Archive>
#> x y dob eol x_id Obj timestamp batch_nr
#> 1: -0.41 3.45 1 NA 1 7.5e-04 2023-09-20 04:41:25.18 1
#> 2: 0.23 -0.79 1 NA 2 5.1e-01 2023-09-20 04:41:25.18 1
#> 3: 1.44 3.39 1 NA 3 2.1e-01 2023-09-20 04:41:25.18 1
#> 4: 3.67 -1.63 2 NA 4 3.4e-07 2023-09-20 04:41:25.20 2
#> 5: 1.96 -0.76 2 NA 5 1.3e-02 2023-09-20 04:41:25.20 2
#> 6: 1.77 -0.94 2 NA 6 1.8e-02 2023-09-20 04:41:25.20 2
s = sel("best")
s$prime(oi$search_space)
mies_survival_comma(oi, mu = 3, survival_selector = s,
n_elite = 2, elite_selector = s)
# sel("best") lets only the best individuals survive.
# mies_survival_comma selects from new individuals (generation 2 in this case)
# and old individuals (all others) separately: n_elite = 2 from old,
# mu - n_elite = 1 from new.
# The surviving individuals have 'eol' set to 'NA'
oi$archive
#> <Archive>
#> x y dob eol x_id Obj timestamp batch_nr
#> 1: -0.41 3.45 1 2 1 7.5e-04 2023-09-20 04:41:25.18 1
#> 2: 0.23 -0.79 1 NA 2 5.1e-01 2023-09-20 04:41:25.18 1
#> 3: 1.44 3.39 1 NA 3 2.1e-01 2023-09-20 04:41:25.18 1
#> 4: 3.67 -1.63 2 2 4 3.4e-07 2023-09-20 04:41:25.20 2
#> 5: 1.96 -0.76 2 2 5 1.3e-02 2023-09-20 04:41:25.20 2
#> 6: 1.77 -0.94 2 NA 6 1.8e-02 2023-09-20 04:41:25.20 2