Description Usage Arguments Details Value Author(s) References See Also Examples
Scales a design of inducing points around a Gaussian measure whose mean is the center of the design matrix and its local neighborhood. The output is an inducing points design centered at the origin that can be used as a template for predictions anywhere in the design space (with a local neighborhood of the same size). Method include scaling by a circumscribed hyperrectangle (chr
) and an inverse Gaussian CDF (qnorm
).
1 2 | scale_gauss_measure_ipTemplate(X, N, gauss_sd, space_fill_design,
method = c('qnorm','chr'), seq_length=20)
|
X |
a |
N |
the positive integer number of Nearest Neighbor (NN) locations used to build a local neighborhood |
gauss_sd |
a vector of standard deviations for the Gaussian measure with with
|
space_fill_design |
a |
method |
the method by which the inducing point template is scaled. In brief,
cHR ( |
seq_length |
an integer that defines the sequence length used to represent the gaussian measure when building the neighbhorhood. |
This function is built to deal with the special class of problems where liGP
is used to predict and integrate over a degenerate Gaussian measure where only
one dimension has a nonzero standard deviation. Separate subroutines are
called for different methods. When method=qnorm
,
qnormscale
is called. The mean of the Gaussian distribution is
the median of the design matrix. The standard deviation of the Gaussian
distribution is one-third of the maximum distance from the median of the
design matrix to the neighborhood points for each dimension.
For each inducing point design, the origin (i.e. predictive location) is
appended to the scaled inducing point design. Thus, the resulting design
contains M+1
inducing points.
The output is a list
with the following components.
Xm.t |
a |
Xn |
a |
Xc |
a |
gauss_sd |
the |
time |
a scalar giving the passage of wall-clock time elapsed for (substantive parts of) the calculation |
D. Austin Cole austin.cole8@vt.edu
D.A. Cole, R.B. Christianson, and R.B. Gramacy (2021). Locally Induced Gaussian Processes for Large-Scale Simulation Experiments Statistics and Computing, 31(3), 1-21; preprint on arXiv:2008.12857; https://arxiv.org/abs/2008.12857
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 | ## Build up a design with N=~40K locations
x <- seq(-2, 2, by=0.02)
X <- as.matrix(expand.grid(x, x))
X_center <- apply(X, 2, median)
## Create inducing point template, first with
## circumscribed hyperrectangle (cHR) scaling
M = 10
Xm <- matrix(runif(2*M), ncol=2)
## Create template with Inverse Gaussian CDF scaling
qnorm_temp <- scale_ipTemplate(X, N=100, space_fill_design=Xm, method="qnorm")
Xm.t_qnorm <- qnorm_temp$Xm.t
Xn <- qnorm_temp$Xn
## Create template with Inverse Gaussian CDF scaling
gauss_sd <- c(0, .05)
qnorm_temp_gauss <- scale_gauss_measure_ipTemplate(X, N=100, gauss_sd=gauss_sd,
space_fill_design=Xm,
method="qnorm")
Xm.t_qnorm_gauss <- qnorm_temp_gauss$Xm.t
Xn_gauss <- qnorm_temp_gauss$Xn
## View different scaled template designs
ylim <- range(Xn_gauss[,2]) + c(-.03, .05)
plot(Xn, pch=16, cex=.5, col='grey',
xlab = 'x1', ylab = 'x2', ylim = ylim,
main='Locally optimized IP template based on Gaussian measure')
points(Xn_gauss, cex=.7)
points(X_center[1], X_center[2], pch=16, cex=1.5)
points(Xm.t_qnorm, pch=2, lwd=2, col=3)
points(Xm.t_qnorm_gauss, pch=6, lwd=2, col=2)
legend('topleft', pch = c(16, 1, 2, 3), col = c('grey', 1, 3, 2),
legend=c('Local neighborhood (qNorm)',
'Local neighborhood (Gauss measure)',
'qnorm ip design',
'Gaussian measure ip design'))
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.