Presentation is loading. Please wait.

Presentation is loading. Please wait.

Basic MC/Defn/Short Answer/Application Cumulative

Similar presentations


Presentation on theme: "Basic MC/Defn/Short Answer/Application Cumulative"— Presentation transcript:

1 Basic MC/Defn/Short Answer/Application Cumulative
Bring a basic calculator Will be given references like Distribution PDF/CDF, and T Table, Chi-Square Table, KS Table Course Outcomes (last 4 definitely represented in long answers)

2 Last 4 Course Outcomes Explain the difference between transient and steady state behavior of a system and how to estimate the steady-state performance parameters such as mean and variance. Compute confidence intervals for a given sample of data and interpret the meaning of the computed confidence intervals. Conduct a goodness-of-fit test using Chi-square hypothesis testing to decide if a given distribution fits a sample of data. Analyze the performance of a single queue or a queuing network using basic queuing theory results

3 Types of Questions Observations
Know properties/definitions so you can compare and contrast them within categories (distributions, similar terminology) Are they similar? What makes them different? Understand process of application. Steps of creating a sim./model. (How was that done in assignments, class example, or new area) How to determine an event probability from a sample space (card games)? More variety of solving questions. Finding mean/variance, determining a distribution, goodness of fit, confidence intervals, Markov chains and queueing theory questions.

4 Modeling (Midterm) What is a model? Limitations?
Approaches? Experimental/Simulation/Analytical. Why choose one for a problem? Model Taxonomy : deterministic, stochastic, static, dynamic, continuous, discrete Terminology: entity, attribute, state, event When not to simulate, Mistakes that can be made Monte Carlo, time-stepped, trace-driven, discrete-event simulation Steps 1,2,3,4,5,6 (goals, conceptual (high level design), specification (details, math, input, pseudocode), computational, verify (model right), validate(right model))

5 Probability/Random Variables/Distributions (Midterm)
Definitions: Random, sample space, probability, independence, exclusive, complementary Continuous random variable: Probability density function PDF Discrete random variable: Probability mass function PMF Cumulative Distribution Function CDF Mean/Variance/Standard Deviation/Coefficient of Variation/Covariance Correlation vs Autocorrelation Discrete vs continuous, infinite vs finite Geometric (events until first succ), Exponential (time between events in Poisson process), Uniform, Poisson (k events in given time) Memoryless property

6 RNG/PRNG (Midterm) What does pseudo-RNG change? What are definitions?
What are desirable mathematical properties? Additional properties for simulation/engineering system? Uniform(0,1)? Statistical testing to show uniformity and independence? Linear Congruential Generator? How does it work? Important features? Period (always full?) Seed Selection, Streams, Mistaken assumptions (complex, unpredictable, seed dangers, system/language specifications effect on accuracy)

7 Random Variate Generation (Midterm)
For discrete (inverse method) Bernoulli trials, uniform, geometric For continuous (inverse method) uniform, exponential Using convolution Geometric, binomial, Poisson Empirical Common Random Variables Discrete -> Uniform, Bernoulli, geometric, binomial, Poisson Continuous -> Uniform, Exponential, Normal

8 Processes (Midterm) Stochastic Process (collection of random variables indexed over time) Counting Process (stochastic process for total # of events in time interval [0,t]) Poisson Process (counting process with a rate lambda) Arrivals vs Inter-arrivals Arrival counts (poisson distribution generated) Inter-arrival times (exponential distribution generated) Pooling property, splitting property Poisson point process (space, not time)

9 Discrete-Event Simulation (Midterm/New)
Concepts activity (defined) vs delay (result of system conditions on run) Components Time-stepping vs Event Scheduling Examples: how to pull out state, entities, events, activities, delays to do design process of modelling Queueing systems Single/multi server, finite/infinite population, tandem queue (blocking), closed network model Service type (FCFS, LCFS, RR, SJF, SRPT) Measures Event List management Array/Linked List/Multiple Linked List/Binary Tree/Heap/Calendar Queue Properties (speed, robustness, adaptability) Choice of data structure

10 Simulation Output Analysis
Transient and Steady State Analysis (Course Outcome) Transient state -> not in steady state (common from initial start of simulation) Steady State -> independent of time and initial start, behaviour over long-run How to collect steady state data while minimizing transient state data Measured variable Y can examine distribution (CDF) statistics like mean and variance Confidence interval of the mean (which is related to the variance) Confidence Intervals and Measuring Error (Course Outcome)

11 Performance Evaluation
Factors – components varied in experiment to see impact on system Levels – settings of the factors Metrics – measurements of the system Experimental Design – Structure of study (method to approach) Vary one factor through levels Vary two factors to see combinations Vary all factors in their levels to explore everything (may not be feasible) Data analysis and presentation

12 Statistical Inference
When you begin from a controlled pRNG seed(s) your single simulation run is in effect a sample path -> results are probabilistic answers to performance evaluation questions -> we can use statistical approaches and methods Chi-Square Testing vs Kolmogorov-Smirnov Testing (Course outcome) Big/Discrete/Buckets vs Small/Continuous/CDF Simulation length – sweet spot of long enough to be reasonable time but results not too influenced by startup transient effects Confidence in results often based around repeated simulations and examination of statistical properties of the results Want consistency/unbiased ANOVA method – analysis of variance (rather than means) to determine what factors most impact the system

13 Simulation Input Analysis
Input drives output of model (garbage in -> garbage out) Desirable to take empirical measurements and reproduce via input model (Such as distribution) Data can be pre-processed but leaving out edge cases can miss realistic biases in input model Checklist (amount, discrete/continuous, range, central tendency, shape, outliers/gaps/anomalies, time series data, correlation, other phenomena) Both questions and visualization or mathematical methods of examining Question for exam? (sample data, what kind of distribution is it?) Once we have a possible solution you can check the properties of the created model against measured data to see if its properties fit as desired (ex QQ plot and goodness of fit tests -> Chi-Square/KS tests (Course Outcome))

14 Queueing Theory Kendall Notation Queuing Systems
Read notation to get properties or create notation based on properties described Defaults Markovian/Deterministic/General Queuing Systems Calling population, system capacity, queue behaviour vs discipline Stability 𝜆<𝑚𝜇, Utilization 𝜌= 𝜆 𝑚 𝜇 and other fundamental rules Little’s Law (mean in system = arrival rate * mean response time)

15 Queueing Theory Markov Models Markov Process
Birth-Death Process, 𝑀/𝑀/1, 𝑀/𝑀/1/𝐾, 𝑀/𝑀/𝑐/𝑐, M/M/∞ Performance (Course Outcome) Markov Chain -> Set of Linear Equations -> State probabilities


Download ppt "Basic MC/Defn/Short Answer/Application Cumulative"

Similar presentations


Ads by Google