Survey
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
1 Pertinent Info/Inputs # of trials N and probability of success p N trials, n different types of success, p1 = probability of type 1 success, p2 = ... , and pn = probability of type n-success. Also, number of success for each type, i.e. ki = # of i successes with the Pn condition i=1 ki = N . Probability of termination p Distribution Binomial (Discrete) Multinomial (Discrete) Geometric (Discrete) We use the Russian Roullette example. An outcome would be death on the third shot. Another outcome is living forever. There are infinitely many outcomes. Different outcomes after N trials, e.g. success of type 1 for all trials. The number of outcomes is kN Different outcomes for N trials, e.g. all successes for all N trials is one outcome in our sample space. The number of elements in our sample space is 2N Ω, the sample space (i.e. the set of outcomes) X(outcome ) = length of the game X(outcome) = (# success of type 1, success of type 2, . . ., success of types n). Note our random variable actually spits out a n-tuple. X(outcome) =# of successes X, the random variable (1) Flip a coin N times (p = 1/2 ) (2)Take a Test with N Questions (probability of answering right p) (3) Meet N people that have a disease (probability of contraction p) (1) have N clients, n workouts and the probability that any client wants to do workout i is pi (2) Have urn of balls colored in n different ways and pick N of them. Also, have the probability is pi for each color i ranging from 1 through n. (1) Russian Roullette with six shooter and spin after each shot (p = 1/6) (2) Quiz show and you stop playing when you get one wrong (probability of getting question WRONG is p) (3) Roll a die until you get a six (p = 1/6) Situations modeled by Random Variable N k « pk (1 − P (X = k) = (p − 1)k−1 p n p1 1 . . . pk n k P (X = (k = „ 1 , . . . , kn )) « N k1 . . . kn p)N −k „ PDF fX (x) x ∈ R or PMF P (X = k) k∈N P (X ≥ k) = (1 − p)k−1 (for Russian. Roulette, means game lasts until k-shot) while F (X) = P (X ≤ k) = 1 − (1 − p)k (means someone gets shot before or on kth turn) Not applicable (1 − p)k−i Continuous CDF F (x) = P (X ≤ x) x ∈ R/ Discrete CDF P (X ≤ k) k∈N „ « Pk N pi i=0 k 1 p We often only want to know the average # of success of type i, in this case we have N pi Np E(X) 1−p p2 Variance of # of success of type i is N pi (1 − pi ). Np(1-p) V(X) In theory, note that the geometric random variable takes on value infinity if the game lasts forever. However, this has probability 0 since P (X = ∞) = limn→∞ pn = 0 as 0 < p < 1. A variant of the geometric distribution is stopping a game after some # of sucesses. Finding the PMF is the content of 12.4.76. If N is very large, then P (X = k) ≈ P (Xλ = k), where Xλ is Poisson random variable with parameter λ and λ is obtained by E(X) = λ = N p. Might be best to not memorize binomial and multinomial, but to be able to think of them combinatorially. See workout example below. Comments 2 Parameter λ. Sometimes you will be given the average lifespan L = E(X) and so λ = 1/L Must be given σ (standard deviation), µ (mean) λ(t) the hazard rate function Normal Distribution (continuous) Hazard-Rate (continuous) interval [a, b] Uniform (Continuous) Exponential (continuous) λ denotes the expected value of success per some fixed interval unit of time. We are also given an interval of time, say t Poisson (Discrete) We use the example of distributions of height. The sample space is the world and each person’s heights. The µ and σ are mean and SD of the heights. Same as exponential We use the lightbulb example. The sample space would be the various times at which the lightbulb could go out once it is plugged in. Sample space would be all the numbers between [a, b] Sample space would be the number of “hits” in some unit time. Note we can have an arbitrarily high number of hits, so our sample space is infinite Same as exponential X(outcome) = the number of people of a particular height X( outcome)= the time when the lightbulb goes out. X(outcome)= the number picked X(outcome) = # of hits (1) Aging (2) More exotic lightbulbs (1) Test Scores (2) Heights (3)Wealth (1) Pick a number on [a, b] at random (2) You arrive at a bus stop between 10 and 11am such that your arrival is uniformly distributed. (1) Radioactive particle decay (2) Lightbulb/ battery durations (1) Patients coming into an ER (λ = patients per hour) (2) Calls into a police station (λ = calls per day). (3) When lifetime of bulb exponentially distributed with parameter λ, and interested # of times we have to replace in some time. Same λ’s for exp. and poisson here! e−λt . S(x) = P (X > x). Can show f (x) = −S 0 (x) when x ≥ 0 and 0 otherwise (on homework using FTC) e 2πσ −(x−µ)2 /(2σ)2 f (x) = √ 1 f (x) = λe−λx when x ≥ 0, and f (x) = 0 when x < 0 1 f (x) = b−a when x in [a, b] otherwise f (x) = 0. (λt)k k! e−λ or if we are given some unit of time t, then we can write: P (N (t) = k) = λk k! P (X = k) = F (x) = 1 − S(t) e √ 2πσ −(x−µ)2 /(2σ)2 −∞ F (x) = 1 − e−λx (for the lightbulb example this means the probability the lightbulb will last at most x units of time) and also S(x) = 1 − F (x) (e.g. the probability the lightbulb will last more than x units of time). F R x(x) = f (x) = R−∞ x 1 x F (x) = b−a when x is in [a, b]. F (x) = 0, when x ≤ a and F (x) = 1 when x ≥ b. P (X ≤ k) = Pk λi −λ i=0 i! e and similarly, for P (N (t) ≤ k). + b) 0 S(t)dt (on homework using integration by parts) R∞ µ 1 λ 1 (a 2 λ − a)2 Not relevant σ 1 λ2 1 (b 12 λ When λ(t) = λ this is just the exponential distribution. 68% occur within one SD, 95% within two SD’s and 99% within 3 SD’s. You use the table after making the substitution u = (x − µ)/σ The Memoryless Property: P (X > t + h|X > t) = P (X > h). Use the lightbulb example to see what this means intuitively. Also, see Poisson Distribution Comments. You computed expectations of max and min’s of independent uniform distributions on your homework Notice that P (N (t) = 0) = e−λt , which is the survival function of an exponetial random variable, thus enabling us to relate the two distributions (see the (3) in situations modeled by exponential). For examples (3), if Tk denotes the waiting time for exponentially distributed random variables, then P (N (t) < k) = P (Tk > t).