Robert B. Gramacy Professor of Statistics
Response Surface Methods and Computer Experiments
STAT 6984 is a graduate "topics" statistics course at the interface between mathematical modeling via
computer simulation, computer model metamodeling (i.e., emulation/surrogate modeling), calibration of
computer models to data from field experiments, and modelbased sequential design and optimization under
uncertainty. The treatment will include some of the historical methodology in the literature, and canonical
examples, but will concentrate on modern statistical methods, computation and implementation in
R
, as motivated by modern application/data type and size.
Notices
 Your final project is posted. It is due 13 Dec, 5pm. A leaderboard is available so you can check your progress relative to your peers.
 The first class is Tuesday, 23 August, 12:301:45pm in RAND 320.
 Classes are canceled the week of Sept 12 (i.e., 13th and 15th).
Lectures

Part 1: Historical Perspective (doc format)

Part 2: Four Motivating Datasets (doc format)
Supplementary code: tpm archive, lockwood archive
Data file(s): lgbb archive, crash archive, and lanl archive

Part 3: Steepest Ascent & Ridge analysis (doc format)
Data file(s): plasma (delta), chemical, rising ridge, saddle point, confidence region 
Part 4: Spacefilling Design (doc format)

Part 5: Basis expansion and splines (doc format)

Part 6: Gaussian process regression (doc format)

Part 7: Design for GPs (doc format)

Part 8: Optimization (doc format)
Auxiliary code file(s): EI example, and toy AL demo.
You'll need an updated laGP package for IECI examples. 
Part 9: Calibration and Sensitivity Analysis (doc format)
Data file(s): wiffle balls 
Part 10: GP Fidelity & Scale (doc format)
You'll need a customized SparseEm package for compactly supported covariances.
Homework Due at the start of lecture
 Homework 1 on regression and optimization, due 6 Sept 2016
Data file(s): wires
Solutions  Homework 2 on steepest ascent, due 15 Sept 2016
Data file(s): sadat, metallurgy, heat transfer, and bumper plating
Solutions  Homework 3 on ridge analysis, due 27 Sept 2016
Data file(s): turbines and viscosity
Solutions  Homework 4 on GPs & splines, now due 18 Oct 2016
Solutions, with code file for surrogate optimization of yield  Homework 5 on design for GPs, now due 27 Oct 2016
Solutions, with code file for LGBB partition modeling (with RData files containing rmses and designs), and EI optimization of yield  Homework 6 on Bayesian optimization, due 10 Nov 2016
Solutions, with code file for the lockwood problem (with RData files containing progress, and IECI optimization of yield  Homework 7 on calibration and sensitivity analysis, due 22 Nov 2016
Solutions  Optional Homework 8 on large scale/high fidelity GP methods, due 8 Dec 2016
Computing
The recommended language for this course is R
,
which can be obtained from CRAN.
Other languages such as MATLAB
are allowed but are not recommended.
Examples in lecture, and help in office hours, etc., will be exclusively in R
.
Below are some helpful R
resources:
 A quick R tutorial and accompanying code file
 Some helpful video tutorials and step by step guides
 R Studio is an excelent multiplatform graphical
interface to
R
which you will likely prefer to the default Windows/OSX GUI(s).