Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Standard view
Full view
of .
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1
Experiments in Bayes Nets

Experiments in Bayes Nets



|Views: 26|Likes:
Published by Adam Smith

More info:

Published by: Adam Smith on Dec 15, 2006
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less





Liu, Smith 1
Jamie Liu and Adam Smith6.825 – Project 211/4/2004We learned a lot from this project. Enjoy.
1. Variable Elimination Functionality
After executing our variable elimination procedure, we obtained the following results for each of the queries below.For the sake of easy analysis of the PropCost probability distributions obtainedthroughout this project from the insurance network, we define the function f to be aweighted average across the discrete domain, resulting in a single scalar valuerepresentative of the overall cost. More specifically,
f = 1E5*P
+ 1E6*P
+ 1E4*P
+ 1E3*P
1. P(Burglary | JohnCalls = true, MaryCalls = true)
<[Burglary] = [false]> = 0.7158281646356072<[Burglary] = [true]> = 0.284171835364393
2. P(Earthquake | JohnCalls = true, Burglary = true)
<[Earthquake] = [false]> = 0.8239331615949207<[Earthquake] = [true]> = 0.17606683840507917
3. P(PropCost | Age = Adolescent, Antilock = False, Mileage = FiftyThou)
<[PropCost] = [HundredThou]> = 0.1729786918964137<[PropCost] = [Million]> = 0.02709352198178344<[PropCost] = [TenThou]> = 0.3427002442093675<[PropCost] = [Thousand]> = 0.45722754191243536(f = 48275.62)
These results are consistent with those obtained by executing the given enumerationprocedure, and those given in Table 1 of the project hand-out.
2. More Variable Elimination Exercise
A. Insurance Network Queries
Liu, Smith 2
1. P(PropCost | Age = Adolescent, Antilock = False, Mileage = FiftyThou,MakeModel = SportsCar)
If the
of the car in question is that of a sports car then,based on the network as illustrated in Figure 1 of the handout, weexpect that the driver would be less risk averse, the driver would havemore money, the car would be of higher value. All of these thingsshould cause the cost of insurance to “go up,” relative to our previousquery which did not involve any evidence about the MakeModel of thecar. An increase in the
domain sense means that theprobability distribution should be shifted towards the higher costelements of the domain (e.g.
might have a higher probabilitythan
).Indeed, this is what happens. As can be seen below,
is about four thousand dollars greater in this case relative to that from Section 1.3.
<[PropCost] = [HundredThou]> = 0.17179333672003955<[PropCost] = [Million]> = 0.03093877334365239<[PropCost] = [TenThou]> = 0.34593039737969233<[PropCost] = [Thousand]> = 0.45133749255661565(f = 52028.74)
2. P(PropCost | Age = Adolescent, Antilock = False, Mileage = FiftyThou,GoodStudent = True)
In this case, counter-intuitive as it may seem, if the driver is a
, then the overall cost of insurance goes up. Thisfollows from the network as shown in Figure 1 of the project handout,i.e.
is only connected to the network through twoparents: Age and
. Since Age is an evidence variable,
is the only node affected by the augmentation of 
to the evidence. More specifically, if the adolescentdriver is a good student, they are likely to have more money, and thusdrive fancier cars, be less risk averse, et cetera.This result is manifested in the results after variable elimination giventhe proper evidence. More specifically,
is a little less than four thousand dollars greater in this case relative to that from Section 1.3.
<[PropCost] = [HundredThou]> = 0.1837467917616061<[PropCost] = [Million]> = 0.029748793596801583<[PropCost] = [TenThou]> = 0.32771416728772235<[PropCost] = [Thousand]> = 0.4587902473538701(f = 51859.40)
Liu, Smith 3
B. Carpo Network Queries
1. P(N112 | N64 = “3”, N113 = “1”, N116 = “0”)
<[N112] = [0]> = 0.9880400004226929<[N112] = [1]> = 0.01195999957730707
2. P(N143 | N146 = “1”, N116 = “0”, N121 = “1”)
<[N143] = [0]> = 0.899999996961172<[N143] = [1]> = 0.10000000303882783
3. Random Elimination Ordering
A. Histograms
Histogram of Computation Time underRandom Elimination Ordering: Problem 1
01000200030004000500060001 2 3 4 5 6 7 8 9 10
Figure 1. Histogram of Computation Time for P(ProbCost | Age =Adolescent, Antilock = False, Mileage = FiftyThou, MakeModel =SportsCar).

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->