genoud_cache: Run the 'genoud' Optimization Function with a Cached Gradient

In libKriging/dolka: Utility functions for Bayesian optimization

Run the genoud Optimization Function with a Cached Gradient

Description

Run the genoud function from the rgenoud package with the objective and the gradient defined in one function. The result is global minimization of a real-valued function of a numeric vector with length d to be provided by fn. The R function fn must provide the values of the objective and its gradient.

The formal arguments of this function and their default values are those of gthe genoud function at the time when this function is written, based on version 5.8.3.0 of rgenoud.

Usage

genoud_cache(
  fn,
  nvars,
  max = FALSE,
  pop.size = 1000,
  max.generations = 100,
  wait.generations = 10,
  hard.generation.limit = TRUE,
  starting.values = NULL,
  MemoryMatrix = TRUE,
  Domains = NULL,
  default.domains = 10,
  solution.tolerance = 0.001,
  gr = NULL,
  boundary.enforcement = 0,
  lexical = FALSE,
  gradient.check = TRUE,
  BFGS = TRUE,
  data.type.int = FALSE,
  hessian = FALSE,
  unif.seed = round(runif(1, 1, 2147483647L)),
  int.seed = round(runif(1, 1, 2147483647L)),
  print.level = 2,
  share.type = 0,
  instance.number = 0,
  output.path = "stdout",
  output.append = FALSE,
  project.path = NULL,
  P1 = 50,
  P2 = 50,
  P3 = 50,
  P4 = 50,
  P5 = 50,
  P6 = 50,
  P7 = 50,
  P8 = 50,
  P9 = 0,
  P9mix = NULL,
  BFGSburnin = 0,
  BFGSfn = NULL,
  BFGShelp = NULL,
  control = list(),
  optim.method = ifelse(boundary.enforcement < 2, "BFGS", "L-BFGS-B"),
  transform = FALSE,
  debug = FALSE,
  cluster = FALSE,
  balance = FALSE,
  ...
)

Arguments

fn	A function to be minimized. This function should return a list with two named numeric elements objective and gradient representing the value of the objective (length 1) and the gradient (length d).
nvars, max, pop.size, max.generations	See genoud.
wait.generations, hard.generation.limit	See genoud.
starting.values	See genoud.
MemoryMatrix	See genoud.
Domains	See genoud.
default.domains	See genoud.
solution.tolerance	See genoud.
gr	Can not be used. This argument is for compatibility
boundary.enforcement	See genoud.
lexical	See genoud.
gradient.check	See genoud.
BFGS	See genoud.
data.type.int	See genoud.
hessian	See genoud.
unif.seed	See genoud.
int.seed	See genoud.
print.level	See genoud.
share.type	See genoud.
instance.number	See genoud.
output.path	See genoud.
output.append	See genoud.
project.path	See genoud.
P1, P2, P3, P4, P5, P6	See genoud.
P7, P8, P9, P9mix	See genoud.
BFGSburnin	See genoud.
BFGSfn	See genoud.
BFGShelp	See genoud.
control	See genoud.
optim.method	See genoud.
transform	See genoud.
debug	See genoud.
cluster	See genoud.
balance	See genoud.
...	Further arguments to be passed to the function given in fn.

Value

The result of a call to genoud.

Caution

This function is hightly experimental and has not been tested enough yet. It should be refactored on the basis of optim_cache which is more achieved.

Note

The optimization is expected to be faster than genoud if the function and its gradient are costly to evaluate separately. However this may not be the case for quite simple functions, due to the cost of setting the "cache" mechanism.

Examples

## Not run: 
## Note that in this example, gradient caching would not be worth it.

dom <- cbind(lower = rep(0.0, 2), upper = rep(1.0, 2))
library(rgenoud)

## emulate a costly-to-evaluate-alone gradient
## ===========================================
braninDer <- function(x) {
   Sys.sleep(0.01)
   branin_with_grad(x)$gradient
}

## separate objective and gradient functions
## =========================================
te <- system.time(res <- genoud(fn = branin, nvars = 2,
                                unif.seed = 123,
                                int.seed = 456,
                                Domains = dom,
                                gr = braninDer))

## gradient "cached"
## ================
teCache <- system.time(resCache <- genoud_cache(fn = branin_with_grad, nvars = 2,
                                                unif.seed = 123,
                                                int.seed = 456,
                                                Domains = dom))
rbind("genoud" = te, "genoud_cache" = teCache)
c("genoud" = res$value, "genoud_cache" = resCache$value)

## Check the use of ...
## ====================
braninShift <- function(x, shift = 0) {
    res <- branin_with_grad(x)
    res$objective <- res$objective + shift
    res
 }
 resShifted <- genoud_cache(fn = braninShift, nvars = 2,
                            unif.seed = 123,
                            int.seed = 456,
                            Domains = dom,
                            shift = 100)
c("genoud_cache" = resCache$value, "genoudShifted" = resShifted$value - 100)

## End(Not run)

libKriging/dolka documentation built on April 14, 2022, 7:17 a.m.