Discrete Random Variable Expectation X ∼ f(x), the expectation (mean or average) of X is x μX =E(X) = Suppose H(x) is a function. Then H(X) is random variable, å x × f (x) o Variance E[H(X)] = å H(X) f (x) x μ = E(X). Var(X) Standard deviation: σX = Var(X) = σ2X = E(X2) − [E(X)]2 Discrete Distribution: PMF, P(X=x) Binomial distribution: o (i). n independent trials; o (ii). each trial has two possible outcomes: success and failure, with probability p and q =(1 – p), respectively; o (iii). X = the total number of ‘success’ in the n trials. æ n ö x n-x ç ÷ p (1- p) è x ø f(x) = , x = 1, 2, 3, …, n o o E(X) = np V(X) = npq, where q = (1—p) Negative Binomial distribution: o (i). independent trials; o (ii). each trial has two possible outcomes: success and failure, with probability p and 1 − p, respectively; o (iii). X = the number of trials needed to produce the rth success. æ x -1 ö r x-r ç ÷ p (1- p) , r -1 ø f(x) = è o o E(X)= r(1-p)/p Poisson Distribution: o Has pmf: o x = r+1,r+ 2, …, n V(X) = r(1-p)/p^2 f (x) = P(X £ x) = l x e- l x! , x = 0,1, 2,..., n o E(X) = λ V(X) = λ Two Discrete RV’s PMF p(x,y) = P(X=x and Y=y) X and Y are independent if: f(x,y)=fx(x) X fY(y) Marginal Probability Mass Function of X & Y, respectively: o for each possible value of x. pX (x) = å p(x, y) y o pY (y) = å p(x, y) for each possible value of y. x Conditional Probability Mass Function of X when Y=y o o f X|Y (x | y) = f (x, y) pY (y) Conditional Probability of X given Y=y: P(X | Y = y) = å f X|Y (x | y) x Expectation: o å åh(x, y)× p(x, y) x o y Conditional Expectation of HX(x) given Y=y: E[H (X) | Y = y] = å H (x)× f X|Y (x | y) x Continuous Random Variable Continuous random variable X takes values in a subinterval of real line R. Definition: The probability density function (pdf) of X is a function given by: o f(x) ≥ 0; ♠ o P(X∈D)= D ò f(x)dx = 1; ò f(x)dx. Definition: The (cumulative) distribution function (df or cdf) of X: x ò f (u)du o F(x)=P(X ≤x)= , o Cdf = F(x) o Pdf = F’(x) = f(x) Expectation: o Definition: X ∼ f(x), the expectation (mean or average) of X is defined by: -¥ m X =E(X)= ò x × f (x)dx s X2 = E(X ) − [E(X)] o Var(X) = o Standard deviation: 2 Percentile: o 0 <= p <= 1 o p = F(η(p)) = ò h ( p) -¥ 2 sX = Var(X) f (u)du Common Continuous Distributions Uniform Distribution: o X follows s uniform distribution on [a,b] with p.d.f. : f(x) = 1/(b—a) , a ≤ x ≤ b E(X) = (a+b)/2 , Var(X) = (b—a)2/12 F(x)=P(X ≤ x)=0, x<a o a≤x≤b 1 b x exp(- ), b o f(x) = x>0 æ xö ç ÷ o F(x) = 1—exp è b ø, x>0 Location Exponential Distribution: o æ x-mö 1 exp ç ÷ b è b ø , x> μ E(X) = f(x) = b +m , V(X) = b2 Double Exponential Distribution: f (x) = o a b 1 -¥ x-a 1 Exponential Distribution: X ∼ Exp(β) l= x ò f (u)du = ò b - adu = b - a, x æ- x -m ö 1 exp ç ÷, 2b è b ø x Î (-¥,¥) E(X) = m , V(X) = 2 / b 2 Normal (or Gaussian) Distribution: f (x) = æ (x - m )2 ö 1 exp ç ÷ 2 2ps è 2s ø o o Standard normal: o Suppose X ~ o o Z= X -m s , x Î (-¥,¥) N(m, s 2 ) ~ N(0,1) P(a <= Z <= b) = , X =sZ +m æb-mö æa-mö Fç ÷ - Fç ÷ è s ø è s ø o æb-mö æa-mö P(X £ a) = F ç ÷ ÷ P(X ³ b) = 1- F ç è s ø è s ø Log-normal Distribution o o log X ~ N(m, s 2 ) æ s2ö E(X) = exp ç m + ÷ è o o 2 ø 2 Var(X) = éëexp(2m + s 2 )ùû[es -1] æ æ ln x - m ö ln x - m ö F(x; m, s ) = P(X £ x) = P [ ln X £ ln x ] = P ç Z £ ÷ = Fç ÷ è è s ø s ø Normal Approximation: o Suppose Y sin Bin(n,p). For np ≥10 and n(1-p) ≥10, the binomial probability for Y can be approximated by a normal distribution with mean np and variance np(1-p). That is, for integers a and b, æ a - 0.5 - np b + 0.5 - np ö P(a £ Y £ b) » P ç £Z£ ÷ npq npq ø è o Gamma Distribution o X ~ Gamma(a, b ) f (x) = o æ xö 1 a -1 x exp ç- ÷ b a G(a ) è bø G(a ) = o where, o E(X) = ab Var(X) = ab 2 o o ò ua -1 -u e du 0 Weibull Distribution a o o o G(k) = (k -1)! ¥ æXö ç ÷ ~ Exp(1) èbø æ 1ö E(X) = b ´ G ç1+ ÷ è aø é æ 2 ö é æ 1 öù2 ù 2 Var(X) = b êG ç1+ ÷ - êG ç1+ ÷ú ú êë è a ø ë è a øû úû ææ X öa æ c öa ö æ æ c öa ö P(X £ c) = P çç ÷ £ ç ÷ ÷ = 1- exp ç - ç ÷ ÷ çè b ø è b ø ÷ ç èb ø ÷ è ø è ø o Beta Distribution o X ~ Beta(α, β) 1 xa -1 (1- x)b -1, B(a, b ) G(a )G(b ) B(a, b ) = G(a + b ) f (x) = o o where, E(X) = o a a+b Var(X) = ab (a + b ) (a + b +1) o Poisson Distribution 2 0 <= x <=1 o f (x) = P(X £ x) = Two Continuous RV’s l x e- l x! PDF P((X,Y ) Î D) = Probability: Expectation: o E[h(X,Y)]= o E(X) = òò ( x,y)ÎD f (x, y)dx, dy ò ò h(x, y) f (x, y)dxdy ò x × f (x)dx x Var(X) = E(X 2 ) -[E(X)]2 Variance: Marginal Distribution of X: f X (x) = o ò f (x, y)dy yÎD Conditional pdf of X given Y=y: f X|Y (x | y) = o f (x, y) fY (y) Conditional Expectation of H(x) given Y=y: ò o x × f X|Y (x | y)dx xÎD Polar coordinate (r,θ) ←→ Cartesian coordinate (x,y) x = r cosθ y = r sin θ dx dy = r dr dθ r^2 = x^2 + y^2 Covariance: o Cov(X,Y) = E(XY) – E(X)E(Y) Correlation: o Corr(X,Y) =r= Cov(X,Y ) s X ×sY Independence: o Multinomial Distribution: o 1.) n independent trials o 2.) each trial has k outcomes the i-th outcome occurs with probability pi i = 1, …, k o 3.) xi = total number of the i-th outcome appeared in the n trials p(x1,…, xn) = P(X1 = x1, X2 = x2, … , Xn = xn) f (x, y) = fX (x)× fY (y) p(x1, … , xr) = n! × p1x1 ×... × prxr (x1 !)...(xr !) Central Limit Theorem: Suppose X1, · · · , Xn are independent with mean μ and variance σ2, for large n, X ̄ approximately follows a normal distribution with μ and variance σ2/n.