30
Importance Sampling

Importance Sampling - Performance Evaluation

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Importance Sampling - Performance Evaluation

Importance Sampling

Page 2: Importance Sampling - Performance Evaluation

What is Importance Sampling ?

AsimulationtechniqueUsedwhenweareinterestedinrareeventsExamples:

BitErrorRateonachannel,Failureprobabilityofareliablesystem

1

Page 3: Importance Sampling - Performance Evaluation

We saw some of it already

Q:We simulate R=10000samples andfind nobiterror.What can we sayaboutthebiterror rate?

A:with confidence0.95,BER<3.710

2

Page 4: Importance Sampling - Performance Evaluation

3

Page 5: Importance Sampling - Performance Evaluation

What is the Problem ?

Assumeyou can simulate asystemYouwant toevaluate theprobability ofarareeventWe want tosay morethan ananswer like : ∈ 0, 3.6910

i.e.we want agoodrelativeaccuracy on

Assumeproba ofrareevent is 10 :howmany simulationruns doyou needtoobtain anestimate of with 10%relativeaccuracy ?

4

Page 6: Importance Sampling - Performance Evaluation

What is the Problem ?

Assumeproba ofrareevent is 10 :howmany simulationruns doyou needtoobtain anestimate of with 10%relativeaccuracy ?

5

Page 7: Importance Sampling - Performance Evaluation

What is the Problem ?

Assumeproba ofrareevent is 10 :howmany simulationruns doyou needtoobtain anestimate of with 10%relativeaccuracy ?

6

replicationsevents

Confidence interval 1.96 1

Relative accuracy = .

1.96 1.96

Relative accuracy = 10% ⇔ 1.96 0.1 ⇔ ..

Page 8: Importance Sampling - Performance Evaluation

The Goal of Importance Sampling

Obtainsmall probability with goodaccuracy…while keeping small

Intheprevious example,thedirectapproach requires 4. 10 runs toestimate 10 with 10%accuracy

We can domuch better with ImportanceSampling

7

Page 9: Importance Sampling - Performance Evaluation

The Idea of Importance Sampling

8

Page 10: Importance Sampling - Performance Evaluation

The Idea of Importance Sampling (cont’d)Ifwe simulate ,howdowe estimate ?

Ifwe simulate insteadofX,we cannot use

But:

Showthis !

9

Page 11: Importance Sampling - Performance Evaluation

Importance Sampling Monte Carlo

10

Page 12: Importance Sampling - Performance Evaluation

Example: Bit Error Rate (BER)

11

Page 13: Importance Sampling - Performance Evaluation

12

Page 14: Importance Sampling - Performance Evaluation

13

, , … ,∼ 0,

discrete, on , , … , ,

, ,Estimate ⋯

1 ⋯

, , … ,on ∞, ∞ discrete, on , , … , ,

Estimate

Page 15: Importance Sampling - Performance Evaluation

14

, , … ,∼ 0,

discrete, on , , … , ,

, ,Estimate ⋯

1 ⋯

, , … ,on ∞, ∞ discrete, on , , … , ,

Estimate

, , ,,

,

Page 16: Importance Sampling - Performance Evaluation

15

Page 17: Importance Sampling - Performance Evaluation

16

, , … ,∼ 0,

12

on ∞, ∞

Page 18: Importance Sampling - Performance Evaluation

17

, , … ,∼ 0,

12

on ∞, ∞

121212

12

∼ ,

Page 19: Importance Sampling - Performance Evaluation

18

, , … ,∼ 0,

discrete, on , , … , ,Estimate ⋯

1 ⋯

, , … ,on ∞, ∞ discrete, on , , … , ,

, , ,

, … ,

Estimate

Page 20: Importance Sampling - Performance Evaluation

Importance Sampling Monte Carlo

Wedothis forseveral valuesof andfindthesame estimate6.4510

What is different ?

19

Page 21: Importance Sampling - Performance Evaluation

Importance Sampling Monte Carlo

Wedothis forseveral valuesof andfindthesameestimate 6.4510What is different ?Hopefully ,thenumber ofruns

20

Page 22: Importance Sampling - Performance Evaluation

Choosing an Importance Sampling Distribution

Whatisagoodimportancesamplingdistribution?Onethatminimizesthenumberofruns

Thiscanbequantifiedwiththevarianceoftheimportancesamplingestimator

21

Page 23: Importance Sampling - Performance Evaluation

22

Page 24: Importance Sampling - Performance Evaluation

Thesmallestvarianceisfor

23

R (proportional to variance)

Page 25: Importance Sampling - Performance Evaluation

Choosing an Importance Sampling Distribution (1)

Ruleofthumb:Theeventsofinterest,undertheimportancesamplingdistributionshouldbe

notrare

notcertain

24

Page 26: Importance Sampling - Performance Evaluation

Choosing an Importance Sampling Distribution (2)

Theoptimalimportancesampling distributionis theonethat minimizes

Isthis thesame asminimizing thevarianceoftheimportancesamplingestimator ?

25

Page 27: Importance Sampling - Performance Evaluation

26

Page 28: Importance Sampling - Performance Evaluation

27

Page 29: Importance Sampling - Performance Evaluation

A Generic Algorithm

Ideas:empiricallyfindimportancesamplingdistributionsuchthatAverageoccurrenceofeventofinterestiscloseto0.5Minimizes

CanbecomputedbyMonteCarlowithsmallnumberofruns

Thealgorithmdoesnotsayhowtodooneimportantthing:whichone?

28

Page 30: Importance Sampling - Performance Evaluation

Conclusion

Ifyouhavetosimulaterareevents,importancesamplingisprobablyapplicabletoyourcaseandwillprovidesiginificantspeedup

Agenericalgorithmcanbeusedtofindagoodsamplingdistribution

29