active_learning {gptools2}

R Documentation

Active learning with surrogate function

Description

Active learning with surrogate function

Usage

active_learning(
  X,
  f,
  surrogate = gp_surrogate$new,
  max_iter,
  tol,
  persistence = 1,
  consecutive = -1,
  parallel = FALSE,
  mc.cores = 1,
  restart,
  options = list(),
  callback,
  ...
)

Arguments

X	A list of evaluation points.
f	A function; the objective function.
surrogate	A function; the surrogate class of object.
max_iter	A positive integer; the maximum iterations.
tol	A numeric vector of length two; the mean-squared-error and out-of-sample-predictive variance tolerance for convergence.
persistence	A positive integer; the number of times the tolerance should be reached before convergence. Use -1 to disable this option.
consecutive	A positive integer; the number of consecutive times the tolerance should be reached before convergence. Use -1 to disable this option.
parallel	TRUE or FALSE; whether to use parallel computing.
mc.cores	A positive integer; the number of cores to use for parallel computing. Only works when 'parallel = TRUE'.
restart	The fitted model from previous run.
options	Optional argument to pass to optim.
callback	Optional function to be called after each iteration of update; this lets users examine the fitting process.
...	Optional arguments to pass to the surrogate function.

Value

A model object returned by the surrogate function.

Examples

## Not run: 
library(gptools2)
s <- seq(-5, 5, length.out = 3)
X <- as.matrix(expand.grid(s, s))
f <- function(x) sin(x[1]) + x[2]

model <- active_learning(
    X, f, sigma = 1e-3,
    max_iter = 100, tol = c(0.1, 0.01)
)

# In-sample fit
pred_y <- predict_gp(model$model, X)
y <- map_row(X, f)
compare(y, pred_y$mean)

# Out-of-sample performance
new_X <- as.matrix(expand.grid(runif(10, -4, 4), runif(10, -4, 4)))
new_y <- map_row(new_X, f)
new_pred_y <- predict_gp(model$model, new_X)
compare(new_y, new_pred_y$mean)

## End(Not run)

---