Functions to supply data to PL for Gaussian process (GP) regression, classification, or combined unknown constraint models
data.GP(begin, end = NULL, X, Y)
data.GP.improv(begin, end = NULL, f, rect, prior,
adapt = ei.adapt, cands = 40,
save = TRUE, oracle = TRUE, verb = 2,
interp = interp.loess)
data.CGP(begin, end = NULL, X, C)
data.CGP.adapt(begin, end = NULL, f, rect, prior,
cands = 40, verb = 2, interp=interp.loess)
data.ConstGP(begin, end = NULL, X, Y, C)
data.ConstGP.improv(begin, end = NULL, f, rect, prior,
adapt = ieci.const.adapt , cands = 40,
save = TRUE, oracle = TRUE, verb = 2,
interp = interp.loess)The output are vectors or data.frames.
positive integer starting time for data to be returned
positive integer (end >= begin) ending time
for data being returned; may be NULL if only data
at time begin is needed
data.frame with at least end rows containing
covariates
vector of length at least end containing real-valued
responses
vector of length at least end containing class labels
function returning a responses when called as f(X)
for matrix X; for data.GP.improv the responses
must be real-valued returned as a vector;
for data.CGP.adapt they must be class
labels returned as a vector;
for data.ConstGP.improv they must be pairs of real-valued
and in {0,1} (1 indicates constraint violation), returned as
a 2-column data.frame
bounding rectangle for the inputs X to f(X) with
two columns and rows equalling nrow(X)
prior parameters passed from PL generated by one of
the prior functions, e.g., prior.GP
function that evaluates a sequential design criterion on
some candidate locations; the default ei.adapt EI
about the minimum; ieci.adapt providing IECI is another
possibility , which is hard coded into data.ConstGP.adapt
number of Latin Hypercube candidate locations used to choose the next adaptively sampled input design point
scalar logical indicating if the improvment information for
chosen candidate should be saved in the psave global variable
scalar logical indicating if the candidates should be
augmented with the point found to maximize the predictive surface
(with a search starting at the most recently chosen input)
verbosity level for printing the progress of improv and other adaptive sampling calculations
function for smoothing of 2-d image plots. The default comes
from interp.loess, but what works best is
interp which requires the interp or
akima package
Robert B. Gramacy, rbg@vt.edu
These functions provide data to PL for Gaussian progress regression
and classification methods in a variety of ways. The simplest,
data.GP and data.CGP supply pre-recorded regression and
classification data stored in data frames and vectors;
data.ConstGP is a hybrid that does joint regression and
classification. The other
functions provide data by active learning/sequential design:
The data.GP.improv function uses expected improvement (EI);
data.CGP.improv uses predictive entropy;
data.ConstGP.improv
uses integrated expected conditional improvement (IECI). In these
cases, once the x-location(s) is/are chosen,
the function f is used to provide the response(s)
Gramacy, R. and Polson, N. (2011). “Particle learning of Gaussian process models for sequential design and optimization.” Journal of Computational and Graphical Statistics, 20(1), pp. 102-118; arXiv:0909.5262
Gramacy, R. and Lee, H. (2010). “Optimization under unknown constraints”. Bayesian Statistics 9, J. M. Bernardo, M. J. Bayarri, J. O. Berger, A. P. Dawid, D. Heckerman, A. F. M. Smith and M. West (Eds.); Oxford University Press
Gramacy, R. (2020). “Surrogates: Gaussian Process Modeling, Design and Optimization for the Applied Sciences”. Chapman Hall/CRC; https://bobby.gramacy.com/surrogates/
PL
## See the demos via demo(package="plgp") and the examples
## section of ?plgp
Run the code above in your browser using DataLab