Simulation optimization means optimizing the performance of a stochastic system that can only be analyzed via simulation. If the number of feasible solutions is finite and small enough, then simulating all of them is possible; ranking & selection (R&S) procedures are algorithms for doing just that in a statistically sound and efficient way. The availability of cheap, high-performance parallel computing has redefined "small enough" to be hundreds of thousands to millions of feasible solution. This is a hands-on class in which students will learn the mathematical-statistics foundations of R&S, apply R&S procedures to simulation optimization problems, and create their own R&S procedure. Knowledge of R and RStudio is useful, but not strictly required, and is only used to allow students to run examples.

R Code

Download these RScript files to participate in the hands-on examples that apply various R&S procedures to five different simulation optimization problems.

• FirstChallenge.R
• Simulations.R
• Procedures.R
• ParallelProcedures.R
  
• MyScript.R (to avoid typing commands)
  

Videos

• Lecture 1: Example, Overview and History (23:40 minutes).
• Lecture 2: Indifference-zone Paradigm (28:07 minutes).
• Lecture 3: Bayesian and Large-Deviation Paradigms (37:43 minutes).
  Erratum: "Ryzhov" is misspelled as "Rzyhov" on slide 23.
• Lecture 4: Subset Selection (14:42 minutes).
• Lecture 5: Fully Sequential Elimination (26:15 minutes).
• Lecture 6: Common Random Numbers (13:28 minutes).
• Lecture 7: Good Selection Guarantees (11:56 minutes).
• Lecture 8: Unknown Variances, KN, and Asymptotic Analysis (20:06 minutes).
• Lecture 9: Parallel R&S Part 1 (36:19 minutes).
  Getting started with doParallel in R.
• Lecture 10: Parallel R&S Part 2 (36:22 minutes).
  
• Lecture 11: Other R&S Problems, Bootstrap R&S, Multi-armed Bandits (23:54 minutes).
  
• R&S Project (4:01 minutes).
  

All slides

References only

Supporting Text

Foundations of Ranking & Selection for Simulation Optimization