UNIT 1_Basic Probability Expt: “randomly choosing a student” All possible outcomes: “any student in the room” Event: “Student chosen is a male” P(A or B) = P(A) + P(B) – P(A and B) {Mutually Exclusive: P(A and B) = 0} P(π΄∩π΅) = P(π΄|π΅)π(π΅) [π(π΄∩π΅) -> fulfil both conditions] [Independent]: P(A|B) = P(A) or P(B|A) = P(B) P(π΄∩π΅) = P(A)P(B) Geologists will predict an earthquake on 95% of occasions for which earthquake occur’ mean P(ElP) or P(PlE) [because earthquake is bound to happen already but its whether geologist can predict] Sensitivity = P(test +ve if disease is present) = P(test +ve | has disease) Positive Predictive Value = P(has disease if tested +ve) = P(has disease | test +ve) Specificity = P(test -ve if disease is absent) = P(test -ve | disease absent) BAYES’ THEOREM π(π΄π|π΅)= "($! ∩&) "(&) P(Ai∩B) = P(B|Ai)P(Ai) If P(A|B) = 0.1, P (C |B) = 0.9 Given 1 – A = C (change the front one) "(B)A*"(&) π(π΄|π΅)= "(+) UNIT 2_(Discrete) Probability Distributions 2.1 Discrete R.V. . E(X) = π, = - π- π₯. Var(X) = π,1 -/0 = - π- (π₯- − π, )1 -/0 2.2 Binomial Distributions 1) n independent trials, 2) each trial has 2 outcomes 3) each trial has same probability - success p, failure 1-p BINOM.DIST (x, n, p, cum) : FALSE - π(π=π₯); TRUE- π(π≤π₯) E(X) = π, = ππ | πππ(π) = π,1 = ππ(1 − π) 2.2 Poisson Distributions 1) Probability of occurrence is the same for any 2 intervals of equal length (time/space) 2) Occurrences in nonoverlapping intervals are independent of one another. POISSON.DIST (x, λ, cum) : FALSE - π(π=π₯); TRUE- π(π≤π₯) X~P(λ) mean = E(X) = λ | πππ(π) = λ parameter λ is avg no. of occurrences per unit time or space (unit time is the time frame they ask you) *No negative in poisson – time scale UNIT 3 – Functions 1 Variable (X): 2 Variables (X&Y): E(aX+bY) = aE(X)+bE(Y) E(aX+b) = aE(X)+b Var(aX+b) = π2Var(X) Var(aX+bY) = a2Var(X)+b2Var(Y)+2abCov(X,Y) a2Var(X)+b2Var(Y)+2abπxπyCorr(X,Y) *applicable to discrete + continuous If X&Y are independent, Cov(X,Y) = 0 Covariance Cov(X,Y) = E[(X-π x)(Y—π y)] ~ Sumproduct(Probabilities,X-u,Y-u) = ∑ π(π = π₯- , π = π¦- )[(π₯- − π, )Eπ₯- − π2 F] Large no. will result in large Cov -> absolute value is not an indicatory of strength or relation Correlation ~ unit free; -1 < corr < 1 [=1] perfectly +ve: linear r/s | [=0] no linear r/s | [=-1] perfect -ve: linear r/s ~ If higher avg values of X are apt to occur with higher avg values of Y, then Cov(X,Y)>0, Corr (X,Y)>0, X & Y are positively correlated. ~ If higher avg values of X are apt to occur w lower avg values of Y, Cov(X,y)<0, Corr(X,Y)<0, X & Y are negatively correlated. ~ Cov(X, Y) = Cov(Y, X) = E(XY)- π, π2 ~ Cov(X, X) = Var(X) Independent Joint Distribution π(π=π₯,π=π¦) = π(π=π₯)π(π=π¦) πΈ(ππ) = πΈ(π)πΈ(π) πΆππ£(π,π) = 0 or πΆπrr(π,π) = 0 -Independent r.v. are always uncorrelated, but dependent r.v. may also be uncorrelated -Linearly related r.v. are perfectly correlated UNIT 4_(Continuous) Probability Distributions for continuous. 3 3 π = ∫43 π₯π(π₯)ππ₯ | π2 = ∫43(π₯ − µ)1 π(π₯)ππ₯ 4.1 Uniform Distributions X is “uniform” on [a,b] if X is equally likely to take any value in the range from a to b. (b>a): X~U[a,b] PDF of X: F(t) = CDF of X: F(t) = E(X) = 0 647 847 647 796 1 if a ≤ t ≤ b, otherwise 0 if a ≤ t ≤ b, otherwise 0 | Var(X) = (647)" 01 4.2 Exponential Distribution Usually for waiting line. Conditions: must occur independently. Occurred in a specific period of time. Must have consistent units. EXPON.DIST(x, π, cum), where π is the rate parameter PDF π(π₯) = ππ−ππ₯ [π₯>0] CDF πΉ(π₯) = 1−π−ππ₯ 0 Mean = πΈ(π)= < 0 πππ(π)= " < In the case of stock out, if mean waiting time per occurrence = 1/3 , π = 3 As such, x= waiting time per occurrence. Since need wait 2 days then restock, x =2 4.3 Normal Distribution π~π(π,π) smaller σ = taller curve, bigger σ = short curve when multiplied with eg n, π=ππ , π=√ππ ~ Symmetric around the mean ~ Highest at its mean NORM.DIST(π,π,π,πππ) NORM.INV(π,π,π) *given p, u, π , find x value) NORM.S.DIST(z, TRUE/FALSE) {π=0, π=1} *given x value, compute p to find u, π NORM.S.INV(p) *given p, compute x value to find u, π Unit 5 Optimisation Excel Modelling: Highlight “Decision Variables” cell [QTY] > Use SUMPRODUCT for “Profit” > “Amt needed for each resources” = SUMPRODUCT -> amt available & amt needed forms constraint. SOLVER-> Set OBJECTIVES = profit > by changing qty > set CONSTRAINTS available qty >= amt needed > SOLVER LP (solving method) UNIT 5: Decision Tree Expected Monetary Value (EMV): weighted average of all possible payoffs for this decision. Expected Value Of Perfect Information (EVPI): EVPI = Max EMV with PI (low risk -> sell, high risk -> no sell) – Max EMV w/o additional info (not perfect) > Must include Cost of i.e Market survey for each relevant payoff at the end Expected Value of Sample Information (EVSI): EVSI = Max EMV with Sample Information (including the cost) + the cost of your survey (negate the cost) – Max EMV without additional information UNIT 6: Linear Programming 1)Decision Variables: each decision variable forms a dimension. n decision variables define an n-dimensional space. A solution is a point in the space. 2)Constraints: each inequality constraint defines a hyperplane(line in a 2D space). Each inequality constraint defines a half space. All constraints collectively define the feasible region. 3)Objective: The objective function defines isoquants and a direction in the space. To finding the optimal soln, push along the direction defined by the objective until we reach the boundary of the feasible region. max π₯# + π₯$ π . π‘. π₯# + 2π₯$ ≤ 3 2π₯# + π₯$ ≤ 3 π₯# , π₯$ ≥ 0 X1+X2=0, then shift, X1+X2=1, X1+X2 =2 (objective function quantity [rhs] ) Slack = max difference in constraint quantity (RHS) before non-binding become binding Infeasible Problem: Empty Feasible region Isoquant: A line on which all points have the same objective value; all points are equally good on the objective function. Optimal solution may not be unique (lie on whole stretch of extreme region), may not be finite(whole feasible region) Redundant Constraint: Add/Remove constraint does not affect feasible region Binding constraints: constraints that are satisfied at equality at the optimal solution. All equality constraints are binding by definition. Non-binding constraints are satisfied at strict inequality at the optimal solution. At optimal solution, some constraints are binding while others are not. The inequality level (=RHS-LHS constraints) is known as the slack. Binding constraints have zero slack by definition. 3 # π· ! π₯!" = π" πππ π = π΄, π΅, πΆ, π· <=> πππ ! ! πππ π₯ππ π=1 π=π΄ !$% UNIT 9: DISCRETE OPTIMIZATION 9.1 Common Uses of Binary Variables All or Nothing can occur π₯1 = π₯0 π₯0 , π₯1 ππππππ¦ For 3 events: π₯0 = π₯1 = π₯= π≤π UNIT 7: SENSITIVITY ANALYSIS If X is not selected, then Z cannot be selected If X and Y are selected, π+π−1≤π then Z must be selected If neither X nor Y is 1−π−π ≤π selected, then Z must be selected At most two of the three π+π+π ≤2 projects can be selected If either X or Y or both is (2 − π − π)/2 ≥ π selected, then Z cannot be selected X is selected only if both Y π ≤ (π + π)/2 and Z are selected Introduce 1/M (Binary P1 will never exceed 1) P1 >= {X11+X12+X13}/M Remember that binary only has 0/1 so if we have inequality value of ½ → 1 B. Change in constraint Changing ππ (constraint quantity) shifts the corresponding constraint. The optimal solution change when constraint is binding. Binding constraints will NOT change if within the sensitivity range (range whereby binding constraint remains as binding constraint). Use the shadow price to estimate the impact on the objective value. Non-binding constraint -> Shadow price 0 Changing πππ (constraint coefficients) rotates the corresponding constraint. Optimal solution, objective value, binding constraint will be affected. No sensitivity range here. Shadow Price: the marginal change of the objective value bcos of an additional unit of resource or requirement. Only valid in the sensitivity range. UNIT 8: OPTIMIZATION Include assumptions, if need be. key components: Decision Variables –[variables can be grouped if there are any complementing patterns] Constraints – Linear, Non-negativity Constraint, Capacity, Resource Constraint, Binary Constraint (Only for Integer Programming). * Simplify expression to just LHS Inequality Sign Fulfilled → >= Max Sold → <= Know the difference between optimal decision/solution (Deliver 800 from plant1 to shop2, 900 to shop4) and optimal objective/value i.e (27200) Standard formation of optimization models Writing out standardization *Always link decision variable with binary variable & i=1,2..5, where M is a large enough number i.e. 1000 OTHER EXCEL FORMULAS SUMPRODUCT(A2:D69,F2:J69) RAND( ), AVERAGE(b11:b41), STDEV.P(b11:b41), VLOOKUP(lookup_value, table array(a2:d66), col no(column from table array), [true/false[exact]), COUNTIF(range, “criteria”), MAX(number 1, number 2…), ROUND(number, 1/2/-1/-2) -1 -> 10th Place -10 -> 100TH place 1 -> 1 dp Lock Excel Formulas $B$4