Package 'plgp'

Title: Particle Learning of Gaussian Processes
Description: Sequential Monte Carlo (SMC) inference for fully Bayesian Gaussian process (GP) regression and classification models by particle learning (PL) following Gramacy & Polson (2011) <arXiv:0909.5262>. The sequential nature of inference and the active learning (AL) hooks provided facilitate thrifty sequential design (by entropy) and optimization (by improvement) for classification and regression models, respectively. This package essentially provides a generic PL interface, and functions (arguments to the interface) which implement the GP models and AL heuristics. Functions for a special, linked, regression/classification GP model and an integrated expected conditional improvement (IECI) statistic provide for optimization in the presence of unknown constraints. Separable and isotropic Gaussian, and single-index correlation functions are supported. See the examples section of ?plgp and demo(package="plgp") for an index of demos.
Authors: Robert B. Gramacy <[email protected]>
Maintainer: Robert B. Gramacy <[email protected]>
License: LGPL
Version: 1.1-12
Built: 2024-11-07 02:55:38 UTC
Source: https://github.com/cran/plgp

Help Index

Particle Learning of Gaussian Processes

Description

Sequential Monte Carlo inference for fully Bayesian Gaussian process (GP) regression and classification models by particle learning (PL). The sequential nature of inference and the active learning (AL) hooks provided facilitate thrifty sequential design (by entropy) and optimization (by improvement) for classification and regression models, respectively. This package essentially provides a generic PL interface, and functions (arguments to the interface) which implement the GP models and AL heuristics. Functions for a special, linked, regression/classification GP model and an integrated expected conditional improvement (IECI) statistic is provides for optimization in the presence of unknown constraints. Separable and isotropic Gaussian, and single-index correlation functions are supported. See the examples section of ?plgp and demo(package="plgp") for an index of demos

Details

For a fuller overview including a complete list of functions, and demos, please use help(package="plgp").

Author(s)

Robert B. Gramacy [email protected]

References

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

Carvalho, C., Johannes, M., Lopes, H., and Polson, N. (2008). “Particle Learning and Smoothing”. Discussion Paper 2008-32, Duke University Dept. of Statistical Science.

Gramacy, R. (2020). “Surrogates: Gaussian Process Modeling, Design and Optimization for the Applied Sciences”. Chapman Hall/CRC; https://bobby.gramacy.com/surrogates/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, tgp

Add data to pall

Description

Add sufficient data common to all particles to the global pall variable, a mnemonic for “particles-all”, for Gaussian process (GP) regression, classification, or combined unknown constraint models

Usage

addpall.GP(Z)
addpall.CGP(Z)
addpall.ConstGP(Z)

Arguments

Z

new observation(s) (usually the next one in “time”) to add to the pall global variable

Details

All three functions add new Z$x to pall$X; addpall.GP also adds Z$y to pall$Y, addpall.CGP also adds Z$c to pall$Y, and addpall.ConstGP does both

Value

nothing is returned, but global variables are modified

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Supply GP data to PL

Description

Functions to supply data to PL for Gaussian process (GP) regression, classification, or combined unknown constraint models

Usage

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)

Arguments

begin

positive integer starting time for data to be returned

end

positive integer (end >= begin) ending time for data being returned; may be NULL if only data at time begin is needed

X

data.frame with at least end rows containing covariates

Y

vector of length at least end containing real-valued responses

C

vector of length at least end containing class labels

f

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

rect

bounding rectangle for the inputs X to f(X) with two columns and rows equalling nrow(X)

prior

prior parameters passed from PL generated by one of the prior functions, e.g., prior.GP

adapt

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

cands

number of Latin Hypercube candidate locations used to choose the next adaptively sampled input design point

save

scalar logical indicating if the improvment information for chosen candidate should be saved in the psave global variable

oracle

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)

verb

verbosity level for printing the progress of improv and other adaptive sampling calculations

interp

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

Details

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)

Value

The output are vectors or data.frames.

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Metropolis-Hastings draw for GP parameters

Description

Functions for using Metropolis-Hastings (MH) to evolve a particle according to the posterior distribution given by a Gaussian process (GP) for regression, classification, or combined unknown constraint model

Usage

draw.GP(Zt, prior, l = 3, h = 4, thin = 10, Y = NULL)
draw.CGP(Zt, prior, l = 3, h = 4, thin = 10)
draw.ConstGP(Zt, prior, l = 3, h = 4, thin = 10)

Arguments

Zt

the particle describing model parameters and sufficient statistics that determines the predictive distribution

prior

prior parameters passed from PL generated by one of the prior functions, e.g., prior.GP

l

positive uniform random walk parameter; for old parameter pold, a new parameter is proposed as p = runif(1, p*l/h, p*h/l). Such proposals are then accepted (or rejected) via the MH acceptance ratio

h

positive uniform random walk parameter; see above

thin

thinning level in the MCMC; describes the number of MH rounds executed before the value is saved as a sample from the (marginal) posterior distribution

Y

not for external use; used internally by CGP and ConstGP internal routines

Details

These functions are used in two important places in plgp. At the user level, they can be used to initialize the particles at time start; see PL and the demos. Internally, they are used in the PL propagate step, e.g., propagate.GP

draw.ConstGP is a combination of the draw.GP and draw.CGP methods, which are for regression and classification GPs, respectively

Value

These functions return an updated particle Zt

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

init.GP, propagate.GP, PL

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

2-d Exponential Hessian Data

Description

Generates 2-d classification data with two or three class labels, based on the Hessian data from a 2-d real-valued response

Usage

exp2d.C(X, threed = TRUE)

Arguments

X

a matrix or data.frame describing the design at which the response categories are desired

threed

a scalar logical indicating if the two or three-class version of the class labels should be returned.

Details

The underlying real-valued response is governed by

Z(X)=x1∗exp⁡(x12−x22).Z(X)=x_1 * \exp(x_1^2-x_2^2).

Two class labels are generated by inspecting the sign of the sum of the eigenvalues of the Hessian (Broderick & Gramacy, 2010). This generates the first (-) and second (+) classes in a three-class function. A third class label (the default) may created from the first one where X[,1] > 0 (Gramacy & Polson, 2011)

Value

A vector of class labels of length nrow(X) is returned

Author(s)

Robert B. Gramacy, [email protected]

References

Broderick, T. and Gramacy, R. (2010). “Classification and categorical inputs with treed Gaussian process models.” Tech. rep., University of Cambridge. ArXiv:0904.4891.

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. (2020). “Surrogates: Gaussian Process Modeling, Design and Optimization for the Applied Sciences”. Chapman Hall/CRC; https://bobby.gramacy.com/surrogates/

https://bobby.gramacy.com/r_packages/plgp/

Examples

## The following demos use this data
## Not run: 
## Illustrates classification GPs on a simple 2-d exponential
## data generating mechanism
demo("plcgp_exp", ask=FALSE)

## Illustrates active learning via entropy with classification
## GPs on a simple 2-d exponential data generating mechanism
demo("plcgp_exp_entropy", ask=FALSE)

## End(Not run)

Initialize particles for GPs

Description

Functions for initializing particles for Gaussian process (GP) regression, classification, or combined unknown constraint models

Usage

init.GP(prior, d = NULL, g = NULL, Y = NULL)
init.CGP(prior, d = NULL, g = NULL)
init.ConstGP(prior)

Arguments

prior

prior parameters passed from PL generated by one of the prior functions, e.g., prior.GP

d

initial range (or length-scale) parameter(s) for the GP correlation function(s)

g

initial nugget parameter for the GP correlation

Y

data used to update GP sufficient information in the case of init.GP; if NULL then pall$Y is used

Value

Returns a particle for internal use in the PL method

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, draw.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Log-Predictive Probability Calculation for GPs

Description

Log-predictive probability calculation for Gaussian process (GP) regression, classification, or combined unknown constraint models; primarily to be used particle learning (PL) re-sample step

Usage

lpredprob.GP(z, Zt, prior)
lpredprob.CGP(z, Zt, prior)
lpredprob.ConstGP(z, Zt, prior)

Arguments

z

new observation whose (log) predictive probability is to be calculated given the particle Zt

Zt

the particle describing model parameters and sufficient statistics that determines the predictive distribution

prior

prior parameters passed from PL generated by one of the prior functions, e.g., prior.GP

Details

This is the workhorse of the PL re-sample step. For each new observation (in sequence), the PL function calls lpredprob and these values determine the weights used in the sample function to obtain the new particle set, which is then propagated, e.g., using propagate.GP

The lpredprob.ConstGP is essentially the combination (product) of lpredprob.GP and lpredprob.CGP for regression and classification GP models, respectively

Value

Returns a real-valued scalar - the log predictive probability

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, propagate.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Extending apply to particles

Description

Applies a user-specified function to each particle contained in the global variables peach and pall, collecting the output in a data.frame

Usage

papply(fun, verb = 1, pre = "", ...)

Arguments

fun

a user-defined function which which takes a particle as its first input; the output of fun should be a vector, matrix or data.frame

verb

a scalar logical indicating whether progress statements should be printed to the screen

pre

an optional character prefix used in the progress print statements; ignored if verb = 0

...

these ellipses arguments are used to pass extra optional arguments to the user-supplied function fun

Details

This is a extension to the built-in apply family of function to particles, intended to be used with the particles created by PL. Perhaps the most common use of this function is in obtaining samples form the posterior predictive distribution, i.e., with the user supplied fun = pred.GP

The particles applied over must be present in the global variables pall, containing sufficient information common to all particles, peach, containing sufficient information particular to each particle, as constructed by PL

Value

Returns a data frame with the collected output of the user-specified function fun

Author(s)

Robert B. Gramacy, [email protected]

References

Carvalho, C., Johannes, M., Lopes, H., and Polson, N. (2008). “Particle Learning and Smoothing.” Discussion Paper 2008-32, Duke University Dept. of Statistical Science.

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, pred.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Extract parameters from GP particles

Description

Extract parameters from particles for Gaussian process (GP) regression, classification, or combined unknown constraint models

Usage

params.GP()
params.CGP()
params.ConstGP()

Details

Collects the parameters from each of the particles (contained in the global variable peach) into a data.frame that can be used for quick summary and visualization, e.g., via hist. These functions are also called to make progress visualizations in PL

Value

returns a data.frame containing summaries for each parameter in its columns

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, lpredprob.GP, propagate.GP, init.GP, pred.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Particle Learning Skeleton Method

Description

Implements the Particle Learning sequential Monte Carlo algorithm on the data sequence provided, using re-sample and propagate steps

Usage

PL(dstream, start, end, init, lpredprob, propagate, prior = NULL,
   addpall = NULL, params = NULL, save = NULL, P = 100,
   progress = 10, cont = FALSE, verb = 1)

Arguments

dstream

function generating the data stream; for examples see data.GP

start

a scalar integer specifying the starting “time”; the data entry/sample where PL will start

end

a scalar integer specifying the ending “time”; the data entry/sample where PL will stop

init

function used to initialize the particles at the start of PL; for examples see draw.GP

lpredprob

function used to calculate the predictive probability of an observation (usually the next one in “time”) given a particle. This is the primary function used in the PL re-sample step; for examples see lpredprob.GP

propagate

function used to propagate particles given an observation (usually the next one in “time”); for examples see propagate.GP

prior

function used to generate prior parameters that may be passed into the dstream, init, lpredprob and propagate functions as needed; for examples see prior.GP

addpall

an optional function that adds the new observation (usually the next one in “time”) to the pall variable in the PL.env environment (i.e., PL.env$pall), which stores the sufficient information shared by all particles; for examples see addpall.GP

params

an optional function called each progress rounds that collects parameters from the particles for summary and visualization; for examples see params.GP

save

an option function that is called every round to save some information about the particles

P

number of particles to use

progress

number of PL rounds after which to collect params and draws histograms; a non-positive value or params = NULL skips the progress meter

cont

if TRUE then PL will try to use the existing set of particles to “continue” where it left off; start and end should be specified appropriately when continuing

verb

if nonzero, then screen prints will indicate the proportion of PL updates finished so far; verb = 1 will cause PL to pause on progress drawings for inspection

Details

Uses the PL SMC algorithm via the functions provided. This function is just a skeleton framework. The hard work is in specifying the arguments/functions which execute the calculations needed in the re-sample and propagate steps.

PL and uses the variables stored in the PL.env environment: pall, containing sufficient information common to all particles, peach, containing sufficient information particular to each of the P particles, and psave containing any saved information. These variables may be accessed as PL.env$psave, for example.

Note that PL is designed to be fast for sequential updating (of GPs) when new data arrive. This facilitates efficient sequential design of experiments by active learning techniques, e.g., optimization by expected improvement and sequential exploration of classification label boundaries by the predictive entropy. PL is not optimized for static inference when all of the data arrive at once, in batch

Value

PL modifies the PL.env$peach variable, containing sufficient information particular to each (of the P) particles

Author(s)

Robert B. Gramacy, [email protected]

References

Carvalho, C., Johannes, M., Lopes, H., and Polson, N. (2008). “Particle Learning and Smoothing.” Discussion Paper 2008-32, Duke University Dept. of Statistical Science.

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

papply, draw.GP, data.GP, lpredprob.GP, propagate.GP, params.GP, pred.GP

Examples

## See the demos via demo(package="plgp"); it is important to
## run them with the ask=FALSE argument so that the
## automatically generated plots may refresh automatically
## (without requiring the user to press RETURN)
## Not run: 
## Illustrates regression GPs on a simple 1-d sinusoidal
## data generating mechanism
demo("plgp_sin1d", ask=FALSE)

## Illustrates classification GPs on a simple 2-d exponential
## data generating mechanism
demo("plcgp_exp", ask=FALSE)

## Illustrates classification GPs on Ripley's Cushings data
demo("plcgp_cush", ask=FALSE)

## Illustrates active learning via the expected improvement
## statistic on a simple 1-d data generating mechanism
demo("plgp_exp_ei", ask=FALSE)

## Illustrates active learning via entropy with classification
## GPs on a simple 2-d exponential data generating mechanism
demo("plcgp_exp_entropy", ask=FALSE)

## Illustrates active learning via the integrated expected
## conditional improvement statistic for optimization
## under known constraints on a simple 1-d data generating
## mechanism
demo("plgp_1d_ieci", ask=FALSE)

## Illustrates active learning via the integrated expected
## conditional improvement statistic for optimization under
## unknown constraints on a simple 1-d data generating
## mechanism
demo("plconstgp_1d_ieci", ask=FALSE)

## Illustrates active learning via the integrated expected
## conditional improvement statistic for optimization under
## unknokn constraints on a simple 2-d data generating
## mechanism
demo("plconstgp_2d_ieci", ask=FALSE)

## End(Not run)

Prediction for GPs

Description

Prediction on a per-particle basis for Gaussian process (GP) regression, classification, or combined unknown constraint models

Usage

pred.GP(XX, Zt, prior, Y = NULL, quants = FALSE, Sigma = FALSE,
        sub = 1:Zt$t)
pred.CGP(XX, Zt, prior, mcreps = 100, cs = NULL)
pred.ConstGP(XX, Zt, prior, quants = TRUE)

Arguments

XX

matrix or data.frame containing (a design of) predictive locations where ncol(XX) = ncol(X), on which the data were trained and particle Zt thus obtained

Zt

the particle describing model parameters and sufficient statistics that determines the predictive distribution

prior

prior parameters passed from PL generated by one of the prior functions, e.g., prior.GP

Y

not for external use; used internally by CGP and ConstGP internal routines

quants

a scalar logical indicating if predictive quantiles should be are desired

Sigma

a scalar logical indicating if the full predictive variance-covariance matrix is desired; typically only used internally by CGP and ConstGP

sub

not for external used; used internally by CGP and ConstGP internal routines

mcreps

number of Monte Carlo iterations used in CGP prediction, integrating over the latent real-valued Y variables at the XX locations

cs

indicates a class label at which the predictive probability is desired; the entire probability distribution over all class labels will be provided if not specified

Details

For pred.GP the predictive mean (and quantiles if quants = TRUE is provided. For pred.CGP the predictive distribution over the class labels is provided, unless only one class (cs) is desired. pred.ConstGP is a combination of the pred.GP and pred.CGP methods

It is suggested that this function is used in as an argument to papply to obtain many predictions - one for each particle in a cloud - which are combined into a data.frame

Some of the function arguments aren't meant to be specified by the user, but are rather there to facilitate usage as a subroutine inside other PL functions, such as lpredprob.GP and others

Value

A single-row data.frame is returned with the desired predictive; these rows are automatically combined when used with papply

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

papply, PL, lpredprob.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Generate priors for GP models

Description

Generate priors for Gaussian process (GP) regression, classification, or combined unknown constraint models

Usage

prior.GP(m, cov = c("isotropic", "separable", "sim"))
prior.CGP(m, cov = c("isotropic", "separable", "sim"))
prior.ConstGP(m, cov.GP = c("isotropic", "separable", "sim"),
              cov.CGP = cov.GP)

Arguments

m

positive scalar integer specifying the dimensionality of the input space

cov

whether to use an "isotropic" or "separable" power exponential correlation function with power 2 – nugget included; a single index model ("sim") capability is provided as “beta” functionality; applies to both regression and classification GPs

cov.GP

specifies the covariance for the real-valued response in the combined unknown constraint GP model

cov.CGP

specifies the covariance for the categorical response in the combined unknown constraint GP model

Details

These function generate a default prior object in the correct format for use with the other PL routines, e.g., init.GP and pred.GP. The object returned may be modified as necessary.

The prior.ConstGP is essentially the combination of prior.GP and prior.CGP for regression and classification GP models, respectively

Value

a valid prior object for the appropriate GP model;

By making the output $drate and/or $grate values negative causes the corresponding lengthscale d parameter(s) and nugget d parameter to be fixed at the reciprocal of their absolute values, respectively. This effectively turns off inference for these values, and allows one to study the GP predictive distribution as a function of fixed values. When both are fixed it is sensible to use only one particle (P=1, as an argument to PL)

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, lpredprob.GP, propagate.GP, init.GP, pred.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

PL propagate rule for GPs

Description

Incorporation of a new data point for Gaussian process (GP) regression, classification, or combined unknown constraint models; primarily to be used particle learning (PL) propagate step

Usage

propagate.GP(z, Zt, prior)
propagate.CGP(z, Zt, prior)
propagate.ConstGP(z, Zt, prior)

Arguments

z

new observation whose to be incorporate into the particle Zt

Zt

the particle describing model parameters and sufficient statistics that the new data is being incorporated into

prior

prior parameters passed from PL generated by one of the prior functions, e.g., prior.GP

Details

This is the workhorse of the PL propagate step. After re-sampling the particles, PL calls propagate on each of the particles to obtain the set used in the next round/time-step

The propagate.ConstGP is essentially the combination of propagate.GP and propagate.CGP for regression and classification GP models, respectively

Value

These functions return a new particle with the new observation incorporated

Author(s)

Robert B. Gramacy, [email protected]

References

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/

https://bobby.gramacy.com/r_packages/plgp/

See Also

PL, lpredprob.GP

Examples

## See the demos via demo(package="plgp") and the examples
## section of ?plgp

Un/Scale data in a bounding rectangle

Description

Scale data lying in an arbitrary rectangle to lie in the unit rectangle, and back again

Usage

rectscale(X, rect)
rectunscale(X, rect)

Arguments

X

a matrix or data.frame of real-valued covariates

rect

a matrix describing a bounding rectangle for X with 2 columns and ncol(X) rows

Value

a matrix or data.frame with the same dimensions as X scaled or un-scaled as appropriate

Author(s)

Robert B. Gramacy, [email protected]

References

https://bobby.gramacy.com/r_packages/plgp/

Examples

X <- matrix(runif(10, 1, 3), ncol=2)
rect <- rbind(c(1,3), c(1,3))
Xs <- rectscale(X, rect)
rectunscale(Xs, rect)