Bayes’ Theorem: P(B)=P(BnA)+P(BnA’) P(A|B)=[P(B|A)P(A)]/[P(B|A)P(A)+P(B|A’)P (A’)] Sensitivity: P(positive|disease) Specificity: P(negative|disease’) PPV: P(disease|positive) Independent Events: P(AnB)=P(A)P(B) Expected Value:∑PiXi Variance: ∑Pi(Xi-µ)2 Binomial Distribution: BINOM.DIST(x,n,p,cumulative) n independent trials; each trial same probability of success and failure E(X)=np, Var=np(1-p) Poisson Distribution:POISSON.DIST(x,λ ,cumulative) probability of occurrence same for any 2 equal intervals; occurrences in non overlapping intervals are independent. P( N=i ) = (e-λxλi /i!) λ = E(X) = mean = Var(X) = average no. of occurrences per unit time Linear Function: Y = aX + b E(Y) = aE(X)+b, Var(Y) = a2Var(X) Cov(X,Y) =∑P(X=Xi,Y=Yi)[(Xi- µx)(Yi- µy)] Cov(X,Y) = E(XY) – E(X)E(Y) Cov(X,X) = Var(X) Corr(X,Y) = (Cov(X,Y)/(σxσy); -1<Corr<1 If independent, E(XY)=E(X)E(Y), Cov=0, Corr=0, P(X=Xi,Y=Yi)=P(X=Xi)*P(Y=Yi) Var(aX+bY)=a2VarX+b2Var(Y), Var(XY)=Var(X)+Var(Y) Random Variables E(aX + b) = aE(X) + b, Var(aX+b) = a2Var(X) E(aX+bY)=aE(X)+bE(Y), Var(aX+bY)=a2VarX + b2VarY + 2abCov(X,Y), PDF: area under curve=1; area = P(a≤X≤b) Mean: ∫xf(x)dx, Var: ∫(x-µ)2 dx CDF: ∫f(s); F(t) = F(<t) Uniform Distribution CDF: f(t) = 1/(b-a) or 0 PDF: F(t) = (t-a)/(b-a) or 0 or 1 E(X): (a+b)/2; Var(X): (b-a)^2/12 Exponential Distribution: EXPON.DIST(x,λ,cumulative) f(x)=λe-λx, x>0 λ=1/E(X); λ2=1/Var(X) F(x) = 1 – e-λx Normal Dist: if X~N(µ,σ), Z=(X-µ)/σ µ-σ<x< µ+σ: 68.26%, µ-2σ<x< µ+2σ: 95.44%, µ-3σ<x< µ+3σ: 99.72% X & Y are normal r.v., W=aX+bY, E(W)=aE(X)+bE(Y)=aµx+bµy Var(W)= a2Var(X)+b2Var(Y) +2ab*σxσyCorr(X,Y) OR a2Var(X)+b2Var(Y)+2abCov(X,Y) Central Limit Theorem: X1,X2,Xn is identically distributed & independent, Sn = X1+X2+Xn (n>30) is approximately normally distributed E(Xi)=µ, E(Sn)=nµ; Var(Xi) = σ2, Var(Sn)=nσ2 Binomial Approx.: if np>5, n(1-p)>5, approx. binomial to normal. X~B(n p,np(1-p)), X~N(np,np(1-p)) Decision Tree: EMV: expected value of all payoffs EVPI: max. value to pay for perfect information, Max. EMV w/ perfect information – max. EMV w/o info. EVSI: max. value to pay for sample info (when info is imperfect) X produces soap. If market strong, will make 18m, if weak, lose 8m. P(strong)=0.3. Note EVSI = 1.36+2.4m. Remember to include cost of survey. Generate U[a,b]: a+(b-a)RAND() Optimisation: remember to define variables clearly. i: Product j: resource Pi: price of product i rj: avail. of resource j; aij: units of resource j needed for one unit of product i max. Woo d F. Hrs C. Hrs Desk 8 Table 6 Chair 1 Avail. 48 4 2 2 1.5 1.5 0.5 20 8 satisfied at equality at optimal solution Sensitivity Analysis: studies the effect of parameter changes, affect optimal solution? Affect optimal objective value? Changes in objective function coeffs.: Max. 40C+50T, s.t. 1C+2T<=10, 4C+3T<=30 Coeff of C [25,67], optimal sol. no change, obj value changes ([25,67] is the sensitivity range of C) To find range of c1 or c2, equate c1/c2=slope=slope of binding variable. Changes in constraint quantity(constr. RHS): 4C+3T<=40, optimal sol.& obj. value changes (draw graphs to see new intersection) within a certain range. when 2 graphs intersect in 1stQ, binding constraints will not change.(shadow price remains) Shadow Price: marginal change of objective value due to +1 unit of resource or requirement (within sensi range) Usually, Yes=1,No=0. e.g. you have $15k, but have 7 investment options (aj), with different NPV (vj) max ∑(j=1, 7) vjxj s.t. ∑(j=1,7) ajxj <=15k, xj =binary At most one event: x1+x2+x3<=1, xi = bin. All or nothing: x1=x2, x1,x2 binary. Conditional: If x1 occurs, x2 must occur x1<=x2, x1,x2 binary If X and Y occurs, Z must occur. X+Y-1<=Z Resolving non-linear expressions: If y does not occur, x=0. If y occurs, 0<=x<=My where y is binary (on/off switch) Decision variables: y1,…,y5: on/off switch x1,…,x5: quantities to produce max ∑(i=1,5) pixi - ∑(i=1,5)riyi Changes in objective function coeff.: e.g. Max Z=4X1+3X2, s.t. 4X1+2X2<=60, 2X1+4X2<=48. Slope of constraint 1: -4/2, constraint 2: -2/4. Thus, 4/2<=C1/3<=-2/4, 1.5<=C1<=6. -4/2<=4/C2<=-2/4, 2<=C2<=8. For shadow price, higher shadow price, better investment. Shadow price: Optimal value (after) – optimal value (before)/change in constraint’s capacity. Objective function is given by final value (what A,B&C should be). The value is 5850 as shown. Note, the table only accounts for one variable to change. If there are 2, e.g. A changes from 50 to 53, and C changes from 55 to 53, take ∑proposed change/allowable change <=100%, if not optimal solution will change. A: (53-50)/7.5 + (55-53)/15 = 0.533 <= 1. So optimal solution will not change. Reduced cost of -7.5 means 1 unit of A would lead to a loss in profit by 7.5. Note also if shadow price <0, when decreased, it will lead to a profit. For questions regarding holding inventory, think Ii-1+Pi=Di+Ii. Thus, min. cost ∑(i=1,6)ciPi+hiIi. Price: Desk $60, Table $30 Chair $20 max. 60x1 + 30x2 + 20x3, s.t. 8x1+6x2+x3<=48, 4x1+2x2+1.5x3 <=20 2x1+1.5x2+0.5x3<=8, x1,x2,x3>=0 Formula: min/max : c1x1+…+cnxn s.t. a11x1+…+a1nxn <= b1, a21x2+…+a2nxn<= b2 Enhancement: max∑(i=1,3)pixi s.t. ∑(i=1,3) aij xj <= rj, for j=1,2,3. Xi>=0, for i = 1,2,3 Binding Constraints: constraints that are e.g. if shadow price of constraint b2 is 6, if b2 is within sensi range, one extra unit of b2 will improve optimal profit by $6. Changes in constraint coeff.: everything changes Allowable increase/decre.:the range, if within, will generate profit as per SP. Discrete Optimisation Int. variables: must be int. Binary variables: must be bin. Mixed Integer Program: mixed bin/int. s.t. ∑(i=1,5)hixi<=4000, ∑(i=1,5)aixi<=4500, xi<=Myi, i=1,..,5 yi binary, xi>=0 Constraint 5: Linda would like to take at least two courses in Marketing (course 11, 18,19, and 20) and at least two courses in Operations (course 4, 14, 15, 16, 17). ๐2,11 + ๐2,18 + ๐2,19 + ๐2,20 ≥ 2๐1,4 + ๐2,4 + ๐1,14 + ๐2,14 + ๐2,15 + ๐2,16 + ๐1,17 ≥ 2 EVSI: -13.9-(-14)+1=1.1, EVPI: -13-(-14)+1=2. If want to find best value of p to maximise EVPI/SI, separate p into bins, place EVSI value beside the header, highlight the table, what-if analysis, column input cell is p’s cell. Maximise 8x1+2x2, s.t. 2x1-6x2≤12, 5x1+4x2≤40 x1+2x2≤12, x1,x2>=0 Xi: quantities to produce Max ∑(i=1,5)pixi - ∑(i=1,5)riyi s.t. ∑(i=1,5) hixi<=4000, ∑(i=1,5) aixi <=4500 0<= xi <= Myi, M is a large number, for i = 1,2,3,4,5. yi binary. ๐๐๐ = {1 ๐๐ ๐ ๐๐ ๐ ๐๐๐๐๐ก๐๐ ๐๐ก ๐ ๐๐ ๐ 0 ๐๐กโ๐๐๐ค๐๐ ๐ Constraint 1: Linda is allowed to take at most five courses in each semester -> X1j <=5 and X2j <=5 Constraint 2: Linda can only take a course if she has completed or is currently taking all courses that are prerequisites for the course (Logic: Those courses with pre-requisite will not be taken in Semester 1) ๐2,9 ≤ ๐1,1 + ๐1,8 − 1 e.g. must take X1,1 and X1,8 before taking X2,9 ๐2,14 ≤ ๐1,4 Must take X1,4 before taking X2,14 Constraint 3: In Semester 1, Linda must take at least three courses from course 1, course 3, course 4, course 5, and course 7. ๐1,1 + ๐1,3 + ๐1,4 + ๐1,5 + ๐1,7 ≥ 3 Then draw graphically to prove . Farmer in Iowa owns 450 of land. Each piece if sell wheat, $2k, 3 workers, 2 fertilisers. If sell corn, $3k, 2 workers, 4 fertilisers. 1000 workers and 1200 fertilisers. Math model: ∑(i=1,2) pixi, s.t. ∑(i=1,2)aijxi<=rj, j=1,2, Xi>=0 Let X be the probability of the amount of ounces poured by the machine into a soda bottle. X ~ N (µ, 0.05) 0.1% of the bottles have to have less than 16 ounces of soda. X~Z(16-µ, 0.05) P(X<16) = 0.001 ((16-µ)/0.05) = -3.09023 16-µ = 0.05*-3.09023 µ = 16 + 0.15451 = 16.15451 Constraint 4: If Linda takes course 18, she will not be allowed to take course 20, because these two courses cover fairly similar materials. ๐2,18 + ๐2,20 ≤ 1 -changing objective function, rotates the isoquant and direction to optimise. Optimal solution will not change if within the sensi range, objective value will always change but we are able to calculate if the optimal solution remains unchanged. -changing coefficient of constraint changes everything. Shifting the constraint will cause optimal solution and objective value to change. Binding constraint will not change if within sensi range. Can use shadow price to estimate the impact on objective value.