MiniMiniMaxUQ-package: Mini-Minimax Uncertainty Quantification for Emulators

In jeff-regier/MiniMiniMaxUQ: Mini-Minimax Uncertainty Quantification for Emulators

Description

MiniMiniMaxUQ seeks to answer the following: How accurately can a black-box function f be emulated from a given set of data? How many evaluations of f are required to guarantee that f can be emulated to a given level of accuracy?

Details

Package:	MiniMiniMaxUQ
Type:	Package
Version:	1.0
Date:	2014-10-20
License:	GPL (>= 2)

Author(s)

Jeffrey Regier and Philip Stark

Maintainer: Jeffrey Regier jeff@@stat.berkeley.edu

References

Jeffrey Regier and Philip Stark. "Mini-Minimax Uncertainty Quantification for Emulators." arXiv preprint arXiv:1303.3079 (2013).

Examples

library(MiniMiniMaxUQ)

X = expand.grid(1:9, 1:9) / 10
f <- function(x) sin(x[1]) + cos(x[2])
f.X = apply(X, 1, f)

K.hat = find.K.hat(X, f.X)

pointwise.uncertainty(X, f.X, K.hat, c(.55,.33))

lower.bound.computational.burden(X, f.X, K.hat, epsilon=.1)
corners.uncertainty.bound(X, f.X, K.hat)
lower.bound.max.uncertainty(X, f.X, K.hat)
upper.bound.max.uncertainty(X, f.X, K.hat)
branch.and.bound.max.uncertainty(X, f.X, K.hat)

uncertainty.confidence.bounds(X, f.X, K.hat)

jeff-regier/MiniMiniMaxUQ documentation built on May 19, 2019, 1:45 a.m.