MUCM-package: MUCM

Description Author(s) References Examples

Description

This package allows the user to estimate the output of a simulator at any point, without actually running it, given a few hundred simulator runs (referred to as the training runs). The emulator essentially determines using Bayesian Analysis the posterior mean and variance of a Gaussian Process for a given input data set, conditioned on the training runs and a user-specified prior mean function. The emphasis is on complex codes that take weeks or months to run, and that have a large number of input parameters; many metrological prediction models fall into this class. A working example is given here for the main functions of this package, which should be the first point of reference.

Author(s)

Sajni Malde, Jeremy Oakley ([email protected]) and David Wyncoll.

References

J. Oakley 1999. 'Bayesian uncertainty analysis for complex computer codes', PhD thesis, University of Sheffield.
Bastos, L. S. and O'Hagan, A. (2009). Diagnostics for gaussian process emulators, Technometrics, 51 (4): 425-438.
Fricker, T. E., Oakley, J. E., & Urban, N. M. (2013). Multivariate Gaussian process emulators with nonseparable covariance structures. Technometrics, 55(1), 47-56.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
# Plot the training data to look for trends
plot(surfebm[1:25, 1:3], lower.panel = NULL)

# Fit the emulator using a linear prior mean (default) and
# a Gaussian correlation function
fit <- fitEmulator(inputs = surfebm[1:25, 1:2], 
                   outputs = surfebm[1:25, 3, drop = FALSE],
                   cor.function = corGaussian)

# Use fitted emulator to predict posterior means and variances at the new points
predictions <- predict(fit, surfebm[26:35, 1:2], sd = FALSE, var.cov = TRUE)

# Compare predictions with true values for the new inputs
# Can also compare accuracy of prediction based on posterior variance
validateEmulator(fit, surfebm[26:35, 3], predictions, plot = TRUE)

OakleyJ/MUCM documentation built on May 7, 2019, 9:01 p.m.