1 / 24

Simulation Analysis and Random Number Generation

DADSS. Simulation Analysis and Random Number Generation. Administrative Details. Homework 6 due Homework 7 due next Monday Very interesting DADSS-like talk taking place tomorrow at noon in the Moot Board Room. Rema Padman

efuru
Télécharger la présentation

Simulation Analysis and Random Number Generation

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. DADSS Simulation Analysis and Random Number Generation

  2. Administrative Details • Homework 6 due • Homework 7 due next Monday • Very interesting DADSS-like talk taking place tomorrow at noon in the Moot Board Room. • Rema Padman • Come! If 5/7 of you come, I’ll include an additional “free” problem on Exam 2

  3. Question • Profit projections • Units sold ~ N(100K,10K) • Unit Price ~ N(25, 6) • Fixed cost = 1mil • Variable cost = N(7.5,7.5) or 2, whichever is greater

  4. What is Simulation? • “A mathematical model that is studied by means of simulation” • Question: Why? • To “simulate” is to replicate or copy the process of something • Many mathematical models are so complicated that it is impossible or impractical to “solve” them via a closed-form or analytic solution

  5. Analytical vs Computational • An example: d = r × t • What happens to distance traveled if I double my speed? • If we want to study the behavior of distance, we have a simple mathematical model with which to do that • Sensitivity of distance with respect to the rate • Sensitivity of distance with respect to the time • Both • Because this model is simple, we can develop an explicit understanding analytically

  6. Analytical vs Computational • Another example: A check-out line in a supermarket • What happens to the amount of time I spend in line if the supermarket opens another register? • Even if we could set out a system of equations that represented this situation (“queueing theory”), it would be very difficult to analyze analytically • Instead, we can analyze the actual problem by building a simulation (or replica) of the problem and studying that • The difference? Control • We get to “experiment”

  7. Possible Future States • Simulation models are useful primarily for allowing examination of what could happen or what could have happened • Two roles for simulation analysis: • Studying uncertainty • Think of a decision tree with 100,000+ of branches • Studying complex systems • Systems for which an analytical representation is impractical or impossible

  8. Simulation Analysis in General • Describe the problem • As always: Choices, States, Acts, Consequences • Create a model of the problem • The role of abstraction • If your simulation model is as complicated as the real thing, why use it? • Provide a characterization of any uncertainty • “Run” the model as many times as necessary to develop an understanding of the system

  9. Building Blocks • Random Numbers • Confidence Intervals • Probability Distributions • Dependence Relationships • Iteration (generally, a “computational mechanism”)

  10. Random Numbers in Theory • What is a random number? • First, note that we can represent any real number just by looking at the range [0,1] • Real number between 0 and 100? • Multiply 100 by RAND() • Real number between -50 and 50? • Multiply 100 by RAND() and subtract 50 • Real number between a and b? • Multiply RAND() by (b - a) and add to a • Only concerned with uniform random numbers between 0 and 1 • everything else can be obtained from uniform [0,1] random numbers • Characteristics of uniform random numbers • Unpredictable • Evenly distributed

  11. Random Numbers in Practice • In Excel: RAND() produces a number that is uniformly distributed between 0 and 1 • That is, each number in that interval is equally likely • True Randomness vs Pseudorandomness • Cycle period (in Excel 1013 random numbers before a cycle) • Prior to Excel 2003, Excel’s cycle used to be 1 million (106)

  12. Using Random Numbers • Back to the coin toss problem • A simple simulation model • Toss a coin until you see one of the following sequences: • HHT: You win $10 • THH: You lose $10 • A coin toss is a random process, so we should be able to build a simulation model to examine this problem • IF(RAND()<0.5,“H”,“T”) • Spreadsheet will be on Blackboard

  13. Coin Toss Results • Solving the coin toss problem by hand, it’s very clear that the answer is that HHT wins 25% of the time and THH wins 75% of the time • However, even running the simulation for 100 iterations doesn’t produce exactly that number as an average result • Simulation models converge on the true solution -- but that true solution is often a limit result (i.e., assuming an infinite number of iterations)

  14. Convergence • In the coin toss example, if we added more runs of the simulation model, our mean answer would get more accurate • The range around it would get tighter • We refer to this process as convergence, and the width of the range around the estimated mean is a confidence interval

  15. Convergence: An Example • Roll a 6-sided die 10 times • How many rolls out of the 10 will result in an even number being thrown?

  16. Convergence: An Example • Roll a 6-sided die 10 times • How many rolls out of the 10 will result in an even number being thrown?

  17. Convergence: An Example • Roll a 6-sided die 10 times • How many rolls out of the 10 will result in an even number being thrown?

  18. The Central Limit Theorem • Convergence is a consequence of the Central Limit Theorem • Simple version: Averages are distributed normally (for all distributions with finite moments) • Consequence: Once we calculate the average of a set of simulation runs, we then know exactly what the uncertainty around that estimated average is because we can calculate the confidence interval

  19. The Central Limit Theorem • 95% Confidence Interval • ALWAYS provide confidence intervals!!! • Convergence is a function of , the square root of the number of runs of the simulation, under most conditions

  20. Speed of Convergence • Although the CLT guarantees convergence, it doesn’t say how quickly it might arrive – it’s a limit property, after all • Often, convergence takes very long – especially for very complicated problems • Convergence occurs on the order of the inverse of the square root of n – the denominator in the confidence interval equation • Can we do better? What about 1/n?

  21. Random Numbers – Without the Randomness • Quasirandom Numbers (“low discrepancy sequences”) • Stratification is what improves convergence, not necessarily randomness • Is randomness critical to the analysis?

  22. Quasirandom Convergence

  23. Quasirandom Numbers • How they work is beyond the scope of this class (it involves number theory) • In general, they are designed to be “self-avoiding” or “space-filling” • By removing randomness completely, you can often obtain better stratification • Variance reduction methods • Latin Hypercube sampling in @Risk • Control Variate Methods • Antithetic Variables Methods • Common Random Numbers • An example to experiment with: Richtmyer numbers (one of the earliest QR sequences)

  24. Imperative! • Simulation is more about doing than about saying! • You must “get your hands dirty” to truly understand this material. • Use the models on Blackboard; build your own • We cannot possibly cover every Excel/@Risk function in class – you will need to explore on your own

More Related