Skip to contents

Locate the next design point for a (D)GP emulator or a bundle of (D)GP emulators using ALM

Source: R/alm.R
alm.Rd

This function searches from a candidate set to locate the next design point(s) to be added to a (D)GP emulator or a bundle of (D)GP emulators using the Active Learning MacKay (ALM), see the reference below.

Usage

alm(object, x_cand, ...)

# S3 method for class 'gp'
alm(object, x_cand, batch_size = 1, M = 50, workers = 1, ...)

# S3 method for class 'dgp'
alm(object, x_cand, batch_size = 1, M = 50, workers = 1, aggregate = NULL, ...)

# S3 method for class 'bundle'
alm(object, x_cand, batch_size = 1, M = 50, workers = 1, aggregate = NULL, ...)

Arguments

object

can be one of the following:

  • the S3 class gp.

  • the S3 class dgp.

  • the S3 class bundle.

x_cand

a matrix (with each row being a design point and column being an input dimension) that gives a candidate set from which the next design point(s) are determined. If object is an instance of the bundle class, x_cand could also be a list with the length equal to the number of emulators contained in the object. Each slot in x_cand is a matrix that gives a candidate set for each emulator included in the bundle. See Note section below for further information.

...

any arguments (with names different from those of arguments used in alm()) that are used by aggregate can be passed here.

batch_size

an integer that gives the number of design points to be chosen. Defaults to 1.

M

[New] the size of the conditioning set for the Vecchia approximation in the criterion calculation. This argument is only used if the emulator object was constructed under the Vecchia approximation. Defaults to 50.

workers

the number of processes to be used for the criterion calculation. If set to NULL, the number of processes is set to max physical cores available %/% 2. Defaults to 1.

aggregate

an R function that aggregates scores of the ALM across different output dimensions (if object is an instance of the dgp class) or across different emulators (if object is an instance of the bundle class). The function should be specified in the following basic form:

  • the first argument is a matrix representing scores. The rows of the matrix correspond to different design points. The number of columns of the matrix equals to:

    • the emulator output dimension if object is an instance of the dgp class; or

    • the number of emulators contained in object if object is an instance of the bundle class.

  • the output should be a vector that gives aggregations of scores at different design points.

Set to NULL to disable the aggregation. Defaults to NULL.

Value

  • If object is an instance of the gp class, a vector is returned with the length equal to batch_size, giving the positions (i.e., row numbers) of next design points from x_cand.

  • If object is an instance of the dgp class, a matrix is returned with row number equal to batch_size and column number equal to one (if aggregate is not NULL) or the output dimension (if aggregate is NULL), giving positions (i.e., row numbers) of next design points from x_cand to be added to the DGP emulator across different outputs. If object is a DGP emulator with either Hetero or NegBin likelihood layer, the returned matrix has two columns with the first column giving positions of next design points from x_cand that correspond to the mean parameter of the normal or negative Binomial distribution, and the second column giving positions of next design points from x_cand that correspond to the variance parameter of the normal distribution or the dispersion parameter of the negative Binomial distribution. If object is a DGP emulator with a Categorical likelihood layer, the returned matrix will have either one column (for binary output) or K columns (for multi-class output), giving the positions of the next design points from x_cand that correspond to the probabilities of different classes.

  • If object is an instance of the bundle class, a matrix is returned with row number equal to batch_size and column number equal to the number of emulators in the bundle, giving positions (i.e., row numbers) of next design points from x_cand to be added to individual emulators.

Details

See further examples and tutorials at https://mingdeyu.github.io/dgpsi-R/.

Note

  • The column order of the first argument of aggregate must be consistent with the order of emulator output dimensions (if object is an instance of the dgp class), or the order of emulators placed in object if object is an instance of the bundle class;

  • If x_cand is supplied as a list when object is an instance of bundle class and a aggregate function is provided, the matrices in x_cand must have common rows (i.e., the candidate sets of emulators in the bundle have common input locations) so the aggregate function can be applied.

References

MacKay, D. J. (1992). Information-based objective functions for active data selection. Neural Computation, 4(4), 590-604.

Examples

if (FALSE) { # \dontrun{

# load packages and the Python env
library(lhs)
library(dgpsi)

# construct a 1D non-stationary function
f <- function(x) {
 sin(30*((2*x-1)/2-0.4)^5)*cos(20*((2*x-1)/2-0.4))
}

# generate the initial design
X <- maximinLHS(10,1)
Y <- f(X)

# training a 2-layered DGP emulator with the global connection off
m <- dgp(X, Y, connect = F)

# generate a candidate set
x_cand <- maximinLHS(200,1)

# locate the next design point using ALM
next_point <- alm(m, x_cand = x_cand)
X_new <- x_cand[next_point,,drop = F]

# obtain the corresponding output at the located design point
Y_new <- f(X_new)

# combine the new input-output pair to the existing data
X <- rbind(X, X_new)
Y <- rbind(Y, Y_new)

# update the DGP emulator with the new input and output data and refit
m <- update(m, X, Y, refit = TRUE)

# plot the LOO validation
plot(m)
} # }