1 1 BN Semantics 1 Graphical Models – 10708 Carlos Guestrin Carnegie Mellon University September 15th, 2008 Readings: K&F: 3.1, 3.2, 3.3 10-708 – ©Carlos Guestrin 2006-2008 10-708 – ©Carlos Guestrin 2008 2 Let’s start on BNs… Consider P(Xi) Assign probability to each xi 2 Val(Xi) Independent parameters Consider P(X1,…,Xn) How many independent parameters if |Val(Xi)|=k?2 10-708 – ©Carlos Guestrin 2008 3 What if variables are independent? What if variables are independent? (Xi ⊥ Xj), 8 i,j Not enough!!! (See homework 1 ) Must assume that (X ⊥ Y), 8 X,Y subsets of {X1,…,Xn} Can write P(X1,…,Xn) = ∏i=1…n P(Xi) How many independent parameters now? 10-708 – ©Carlos Guestrin 2008 4 Conditional parameterization – two nodes Grade is determined by Intelligence3 10-708 – ©Carlos Guestrin 2008 5 Conditional parameterization – three nodes Grade and SAT score are determined by Intelligence (G ⊥ S | I) 10-708 – ©Carlos Guestrin 2006-2008 6 The naïve Bayes model – Your first real Bayes Net Class variable: C Evidence variables: X1,…,Xn assume that (X ⊥ Y | C), 8 X,Y subsets of {X1,…,Xn}4 10-708 – ©Carlos Guestrin 2006-2008 7 What you need to know (From last class) Basic definitions of probabilities Independence Conditional independence The chain rule Bayes rule Naïve Bayes 10-708 – ©Carlos Guestrin 2006-2008 8 This class We’ve heard of Bayes nets, we’ve played with Bayes nets, we’ve even used them in your research This class, we’ll learn the semantics of BNs, relate them to independence assumptions encoded by the graph5 10-708 – ©Carlos Guestrin 2006-2008 9 Causal structure Suppose we know the following: The flu causes sinus inflammation Allergies cause sinus inflammation Sinus inflammation causes a runny nose Sinus inflammation causes headaches How are these connected? 10-708 – ©Carlos Guestrin 2006-2008 10 Possible queries Flu Allergy Sinus Headache Nose Inference Most probable explanation Active data collection6 10-708 – ©Carlos Guestrin 2006-2008 11 Car starts BN 18 binary attributes Inference P(BatteryAge|Starts=f) 218 terms, why so fast? Not impressed? HailFinder BN – more than 354 = 58149737003040059690390169 terms 10-708 – ©Carlos Guestrin 2006-2008 12 Factored joint distribution - Preview Flu Allergy Sinus Headache Nose7 10-708 – ©Carlos Guestrin 2006-2008 13 Number of parameters Flu Allergy Sinus Headache Nose 10-708 – ©Carlos Guestrin 2006-2008 14 Key: Independence assumptions Flu Allergy Sinus Headache Nose Knowing sinus separates the symptom variables from each other8 10-708 – ©Carlos Guestrin 2006-2008 15 (Marginal) Independence Flu and Allergy are (marginally) independent More Generally: Flu = t Flu = f Allergy = t Allergy = f Allergy = t Allergy = f Flu = t Flu = f 10-708 – ©Carlos Guestrin 2006-2008 16 Conditional independence Flu and Headache are not (marginally) independent Flu and Headache are independent given Sinus infection More Generally:9 10-708 – ©Carlos Guestrin 2006-2008 17 The independence assumption Flu Allergy Sinus Headache Nose Local Markov Assumption: A variable X is independent of its non-descendants given its parents and only its parents (Xi ⊥ NonDescendantsXi | PaXi) 10-708 – ©Carlos Guestrin 2006-2008 18 Explaining away Flu Allergy Sinus Headache Nose Local Markov Assumption: A variable X is independent of its non-descendants given its parents and only its parents (Xi ⊥ NonDescendantsXi | PaXi)10 10-708 – ©Carlos Guestrin 2006-2008 19 What about probabilities? Conditional probability tables (CPTs) Flu Allergy Sinus Headache Nose 10-708 – ©Carlos Guestrin 2006-2008 20 Joint distribution Flu Allergy Sinus Headache Nose Why can we decompose? Local Markov Assumption!11 10-708 – ©Carlos Guestrin 2006-2008 21 A general Bayes net Set of random variables Directed acyclic graph CPTs Joint distribution: Local Markov Assumption: A variable X is independent of its non-descendants given its parents and only its parents – (Xi ⊥ NonDescendantsXi | PaXi) 10-708 – ©Carlos Guestrin 2006-2008 22 Announcements Homework 1: Out wednesday Due in 2 weeks – beginning of class! It’s hard – start early, ask questions Collaboration policy OK to discuss in groups Tell us on your paper who you talked with Each person must write their own unique paper No searching the web, papers, etc. for answers, we trust you want to learn Audit policy No sitting in, official auditors only, see couse website Don’t forget to register to the mailing list at: https://mailman.srv.cs.cmu.edu/mailman/listinfo/10708-announce12 10-708 – ©Carlos Guestrin 2006-2008 23 Questions???? What distributions can be represented by a BN? What BNs can represent a distribution? What are the independence assumptions encoded in a BN? in addition to the local Markov assumption 10-708 – ©Carlos Guestrin 2006-2008 24 Today: The Representation Theorem – Joint Distribution to BN Joint probability distribution: Obtain BN: Encodes independence assumptions If conditional independencies in BN are subset of conditional independencies in P13 10-708 – ©Carlos Guestrin 2006-2008 25 Today: The Representation Theorem – BN to Joint Distribution If joint probability distribution: BN: Encodes independence assumptions Obtain Then conditional independencies in BN are subset of conditional independencies in P 10-708 – ©Carlos Guestrin 2006-2008 26 Let’s start proving it for naïve Bayes – From joint distribution to BN Independence assumptions: Xi independent given C Let’s assume that P satisfies independencies must prove that P factorizes according to BN: P(C,X1,…,Xn) = P(C) ∏i P(Xi|C) Use chain rule!14 10-708 – ©Carlos Guestrin 2006-2008 27 Let’s start proving it for naïve Bayes – From BN to joint distribution Let’s assume that P factorizes according to the BN: P(C,X1,…,Xn) = P(C) ∏i P(Xi|C) Prove the independence assumptions:
View Full Document