ECE 450 – Lecture 2 • Recall: Pr(A B) = Pr(A) + Pr(B) – Pr(A B) in general = Pr(A) + Pr(B) if A and B are m.e. Lecture Overview – Conditional Probability, Pr(A | B) – Total Probability – Bayes’ Theorem – Independent Events – Compound Experiments – Binomial Distribution ECE 450 D. van Alphen 1 Conditional Probability • Defn: The conditional probability of event A, given that event B has occurred is Pr( A | B) Pr( A B) , Pr(B) P(B) 0 (1) • Note (by symmetry of the definition): Pr( A B) Pr(B | A ) Pr( A ) P( A ) 0 (2) • From (1) & (2), we have two new ways of writing Pr(A B): Pr(A B) = Pr(A) Pr(B | A) = Pr(B) Pr(A | B) ECE 450 D. van Alphen 2 Verification Example (Is the definition reasonable?) • Experiment: Toss a single die, and find Pr(2 | even) Pr(2 | even) Pr(2 even) Pr(2) 16 1/ 3 Pr(even) Pr(even) 3 6 • Another way to look at it - Let B be the event of getting an even number: B is called the restricted S B’ B sample space; 2 is now 1 4 5 2 one of 3 equally likely outcomes. 3 6 ECE 450 D. van Alphen 3 Another Example • Experiment: Toss 2 coins • Sample Space: S = {HH, HT, TH, TT} • Find the conditional probability of obtaining two heads when flipping a coin twice, given that at least one head was obtained; i.e., find Pr(2 heads | at least 1 head) • Def: A: event of obtaining 2 heads = {HH} B: event of obtaining at least one head = {HH, HT, TH} • Then Pr( A B) Pr( A ) 1 4 Pr( A | B) 1/ 3 Pr(B) Pr(B) 3 4 ECE 450 D. van Alphen 4 Side Note & Definition • Note: Conditional probabilities are themselves probabilities; thus, they satisfy all the axioms for probabilities: 1. Pr(A | B) 0 2. Pr(B | B) = 1 3. Pr(A C | B) = Pr(A|B) + Pr(C|B) if A and C are m.e. • Another definition: consider a collection of subsets, {Ai}, (i = 1, …, n ), of S. The collection is S A A 2 1 a partition of S if: ... Ai Aj = f, i j An U Ai = S, i = 1, …, n ECE 450 D. van Alphen 5 Total Probability Theorem • Let {Ai} be a partition of S, and let B be a subset of S: S A1 A2 A3 B ... An Strategy: To find Pr(B), break apart B, into mutually exclusive pieces • Then Pr(B) = Pr[ (A1 B) (A2 B) . . . (An B) ] m.e = Pr(A1 B) + Pr(A2 B) + … + Pr(An B) (see box, bottom of p.2) Pr(B) = Pr(B|A1)Pr(A1) + Pr(B|A2)Pr(A2) + … + Pr(B|An)Pr(An) ECE 450 D. van Alphen 6 Example Using “Total Probability” Transistor types X , Y and Z make up 45%, 25%, and 30% of the total number of transistors in a box, respectively. Let B be the event that a transistor fails before 1000 hrs. Given the reliability information: Pr(B|X) = .15 Pr(B|Y) = .4 Pr(B|Z) = .25 Find the probability that a randomly chosen transistor from the box fails before 1000 hours. Pr(B) = ______ _____ + ________ ____+ _______ _____ = _____ ______ + ______ ____+ ______ ______ = .243 ECE 450 D. van Alphen 7 Bayes’ Rule • Recall from p. 2 (again, the box near the bottom) Pr(A B) = Pr(A|B) Pr(B) = Pr(B|A) Pr(A) Focus here; solve for Pr(A|B) Pr(B | A ) Pr( A ) Pr( A | B) , Pr(B) Pr(B) 0 (Bayes’ Rule) Pr(B | A j ) Pr( A j ) Pr( A j | B) n Pr(B | A j ) Pr( A j ) i1 ECE 450 D. van Alphen By __________ _____________ 8 Bayes’ Rule: Note & Example • Note: Use Bayes’ Rule when asked to find some Pr(A|B), but it would be easier to find Pr(B|A). (“Backwards conditional probability”) Example: Say an observed transistor (from the previous example) fails before 1000 hrs. Find the probability that it was a Type Z transistor. Let B: event that transistor fails before 1000 hrs. We want: Pr(Z|B), non-trivial We know Pr(B|Z) = .25 (the easier problem; given on p. 7) Using Bayes’ Rule, next page: ECE 450 D. van Alphen 9 Bayes’ Rule Example, continued Pr(B | Z) Pr( Z) Pr( Z | B) Pr(B) .30864 Answer from p. 7 example In-Class Practice • Two cards are drawn without replacement from a 52-card deck. • Find the probability that the 2nd is a queen, given that the 1st is a queen. • Find the probability that both the 1st and 2nd are queens. • Unordered answers: 1/221, 3/51 ECE 450 D. van Alphen 10 Bayes’ Rule In-Class Example A medical test for a particular type of cancer has the following properties: • It correctly detects the cancer (when present) with probability 95%; • It incorrectly “detects” the cancer (when there is no cancer present) with probability 20%. Suppose that this particular type of cancer is present in only 1% of people of your age/sex/ethnicity. Find the probability that you actually have this cancer, given that your test is positive (i.e., cancer was detected). Let c: event that you have cancer. Let d: event that cancer is detected. ECE 450 D. van Alphen 11 Pr(d|c) = .95; Pr(___|___) = .05 (complements) Pr(d|c’) = .2; Pr(d’|c’) = _______; Pr(c) = .01 (a priori) Find Pr(c|d) _______________________________________________ _______________________________________________ _______________________________________________ _______________________________________________ _______________________________________________ _______________________________________________ _______________________________________________ _______________________________________________ _______________________________________________ ECE 450 D. van Alphen 12 (Statistically) Independent Events • Defn: 2 events A and B are statistically independent (__) if and only if Pr(A|B) = Pr(A) (knowing B has occurred tells me nothing about whether or not A has occurred) • Equivalently: Pr(B|A) = P(B) and Pr(A B) = Pr(A) Pr(B) (caution: only true for independent events) ECE 450 D. van Alphen 13 Example Using Independence Definition • Experiment: Toss 2 dice. Define the events A: sum = 7 Question: Are A, B, B: 1 face = 6 • Pr(A | B) Pr Pr(B) Pr{(6,1) (1, 6)} 2 / 36 2 Pr{(6,1) (6, 2) (6, 6)} 11/ 36 11 A, B not _____ • Pr(A) = 6/36 = 1/6 A, B dependent ECE 450 D. van Alphen 14 Facts/Thoughts About Independence • Fact 1: Complementary events are dependent events. • Fact 2: Unions, intersections & complements of independent events are independent. • Consider the 2012 election (Obama/Biden vs. Romney/Ryan): Is there any pair of – Event A: Obama wins as Pres. events from this set – Event B: Romney wins as Pres. that is an independent – Event C: Biden wins as VP. pair? – Event D: Ryan wins as VP. • Let A: it rains today; B: it rains tomorrow. Are A and B independent ? ECE 450 D. van Alphen 15 Example Using Independence • Say the switches in the circuit below are closed ( -ly of each other) at any time with probability 0.1. Find the probability of a closed path from point A to point B. B A • Labeling for the Solution: top right B A bottom ECE 450 D. van Alphen Note: for a closed path to exist, the “right” switch must be closed; and, either the “bottom” switch must be closed or both of the “top” switches must be closed. 16 Example, continued Pr(closed path) = Pr[‘right’ and (‘top’ or ‘bottom’)] (both switches) = Pr(right) Pr(top or bottom), by Fact 2, p. 14 = (0.1) [Pr(top) + Pr(bottom) – Pr(both top & bottom)] by Corollary 4 (both switches: (.1)(.1) ) = .1 [(.01) + .1 – (.01)(.1)] = .l0109 top right Note: Don’t round off! B A ECE 450 bottom D. van Alphen 17 Compound Experiments • Consider experiments: Ea, Eb with sample spaces: Sa, Sb • Say Sa = {a1, …, an} and Sb = {b1, …, bm}. • Define Sa x Sb (the Cartesian Cross Product) as the set of ordered pairs with the 1st element from Sa and the second element from Sb • Do experiments Ea & Eb (jointly), and get pairs of outcomes from Sa x Sb • The joint performance of Ea & Eb is said to be a compound experiment. ECE 450 D. van Alphen 18 Cross Product Examples & Bernoulli Trials Example 1: Toss a coin 2 times, with S1 = S2 = {H, T}. S = S1 x S2 = { ____, ____, _____, _____} Example 2: Toss a coin 3 times, with S1 = S2 = S3 = {H, T}. S = S1 x S2 x S3 = { (HHH), (HHT), (HTH), (HTT), (THH), (THT), (TTH), (TTT) } Definition: A Bernoulli Trial is an experiment with only 2 possible outcomes, sometimes called “success” and “failure”. Success 1 yes, ECE 450 Failure 0 no D. van Alphen 19 Binomial Experiments • Definition: A Binomial Experiment is an experiment consisting of n independent Bernoulli Trials. – Let Ak denote the event of getting k successes (and thus n-k failures) in n trials. – Let p be the probability of success on each trial. – Let q = 1 – p be the probability of failure on each trial. • Example: Consider a 5-trial binomial experiment, and say we are interested in the probability of having 2 successes (first), followed by 3 failures. – Pr{1, 1, 0, 0, 0} = __ __ __ __ __ = ______ (O.K. to multiply since the events are ____) ECE 450 D. van Alphen 20 Binomial Experiments, continued • Refined example: Say we want to know the probability of getting 2 successes, in 5 independent trials (order doesn’t matter): 5 2 3 Pr{A2} = p q 2 (since there are 5C2 ways to decide where to put the 2 successes in the string of 5 outcomes) • In general, the probability of getting k successes in n (independent) Bernoulli tials is n k n – k pn(k) = Pr{Ak} = p q k ECE 450 D. van Alphen (Binomial Experiments) 21 Binomial Experiment Examples 1. 10 missiles are fired at a tank; each missile has a (0.2) probability of hitting the tank, independently of the other missiles. Find the probability that exactly 3 missiles hit the tank. p10(3) = Pr{ A3} = (__)3 (__)7 .2013 2. A certain football player can catch 2/3 of the passes thrown to him. He needs to catch at least 3 more passes for his team to win the game. Find the probability that his team wins if the quarterback throws to him 5 more times. Pr{win} = Pr{at least 3 catches} = Pr{exactly 3 catches or exactly 4 catches or exactly 5 catches} ECE 450 D. van Alphen 22 Binomial Experiment Examples, continued = Pr{A3 A4 A5} = Pr{A3} + Pr{A4} + Pr{A5} = ___ ___ ___ + ___ ___ ___ + ___ ___ ___ = .790123 ** Note: we can add these probabilities, because the events “catch exactly 1 pass”, “catch exactly 2 passes”, and “catch exactly 3 passes” are: _____________________________ events. ECE 450 D. van Alphen 23 Binomial Experiment Examples 3. Using a normal deck of cards, say we cut the deck 5 times; find the probability of getting an ace on at least 3 cuts. Pr{at least 3 aces} = Pr{exactly 3 aces or exactly 4 aces or exactly 5 aces) (Bernoulli Trials) m.e. = Pr{A3} + Pr{A4} + Pr{A5} 5 1 3 12 2 5 1 4 12 1 5 1 5 12 0 = 3 13 13 4 13 13 5 13 13 1501 = ECE 450 135 .00404 D. van Alphen 24 Danger!! 1. Pr(A B) = Pr(A) + Pr(B) – Pr(A B) = Pr(A) + Pr(B) if A, B m.e. 2. Pr(A B) = Pr(A) Pr(B | A) = Pr(A) Pr(B) ECE 450 if A, B independent D. van Alphen 25 Review • Pr(A|B) = ___________________ (defn) = ___________________ (Bayes’ Rule) • Total Probability: If {Ai} is a partition of sample space S, then Pr(B) = ___________________________________________ • If A and B are independent, then Pr(A|B) = ______________ and Pr(A B) = ____________ • (Binomial Experiments): The probability of getting k successes in n independent Bernoulli trials is: _____________________________________ ECE 450 D. van Alphen 26