Maximin Criterion
based of the worst-case scenario
Calculate the minimum payoff of all choices and select the one that perform best.
Minimax Regret Criterion
The payoff table is based on “lost opportunity”
Calculate the lost opportunity for each entry in the table (the maximum gain per state - what that choice actually got).
Create a new State called regret, by selecting the max regret for each action and select the action that is there the smallest.
Maximax Criterion
This criterion is based on the best possible scenario
Calculate the maximum payoff of all choices and select the one that perform best.
Principle of Insufficient Reasoning
Calculate the net gain per choice, assuming that all choices are weight the same, and select the one that perform best.
Decision making criterion under risk
different outcomes have probabilities that state how likely it is to occur
Decision Making Under Uncertainty
Decision making without probabilities
The Expected Value Criterion (Bayes rule)
Hodges-Lehmann
Calculated by getting the weighted sum of all states (EP) (by multiplying it with the probability) and the minimum (min)
Expected Return with Perfect Information
best outcome per state, weight by likelyhod
Expected Return under Risk/Expected Return of the Expected Value criterion
Sum all weighted cells per action, select the action’s value that performs best
Expected Value of Perfect Information
(Expected Return with Perfect Information) - (Expected Return under Risk)
Expected payoff with forecast
Efficiency
(Expected payoff with forecast) / (Expected Value of Perfect Information)
Linear Programming
Ein Problem ist linear,
Wenn die Zielfunktion linear ist
Wenn alle Nebenbedingungen linear sind
Aufbau eines LPs
Variablen definieren
Zielfunktion definieren
constraints definieren
Nicht-0 Bedingung
Gelöst graphisch
Sensitivitätsanalyse und Schattenpreis
Eine Nebenbedingung wird verändert
Schattenpreis = Veränderung der optimalen Lösung (pro Einheit Nebenbedingung)
Transportproblem
When Supply > Demand
Create a new demand location b_{J+1}, to fill demand. (all cost beeing 0)
When Demand > Supply
Create a new supply location b_{J+1}, to fill supply. (all cost beeing ???)
Northwest Corner Method
Oben links maximalen Wert geben
Nächstmögliche Zelle maximieren
Repeat
Always yields solution
Minimum‐Cost Method
Find min ship cost, and fill cell
is greedy
Vogel’s Approximation Method
penalty cost = secondMin(row/column) - min(row/column)
Find row/column with largest penalty
Fill min cost in that row
Recompute all penalties
heuristic
Stepping Stone Method
Start with a feasible solution
for each empty cell, check if its value +1 better (see circle)
If better, do it, till the first cell gets to 0
The Assignment Problem
Divide a number of task between a number of people (each task can only be done by one person).
⇒ Special Type of Transportation Problem (Demand = 1 & supply = 1)
Each Task can’t be partially done / cant be split
The Hungarian Method
Find minimal value in each row (not 0) and subtract every item in that row with that value
Find minimal value in each column (not 0) and subtract every item in that row with that value
Can you put x’es on a subset of all 0rors, so that each column and row has exactly 1 If posible go to 5.
Find the minimum value, not covert by these lines, and substract that from every uncovert value. And add 30 to all points where the lines intersect. Repeat to 3
Then fill these x’es in. Those are the solutions.
The Knapsack Problem
You have a set of items, each with its own value and weight. What items do you take, considering your max capacity and wanting max value.
Profitability Index (Knapsack)
Calculate Profitability index, value / weight
Pick highest until full
Dynamic Optimization
for i in row:
for k in column:
if i = 0:
// Keine Items zur auswahl
set 0
else:
// die Items 1 - i stehen zur auswahl
// Maximiere diesen Wert in dem optional:
// ein wert aus der unteren reihe
// oder den eigenen wert genommen wird
kapa = k
for i in reverse(row):
if A[i-1][kapa] = A[i][kapa]: // Element darunter
// Bedeutet Element nicht im Sack
// Element im Sack
kapa = kapa - w[i]
Dijkstra‘s Algorithm
Nur für positive Gewichte
Each node gets a label [distance; vorgänger]
Start with item with smallest distance
Assign to be permanent
Recalc all reachable nodes labels
Minimum Spanning Tree (MST)
Kleinster Tree der alle Nodes einmal berührt.
Prim’s Algorithm
Con := Set all connected
Uncon := Set all not connected
zufällige node $v$
Con := $\{v\}$
Uncon := $V \setminus v$
Find $e \in Uncon$ that is next to (lowest weight) a Con weight. Add that note to Con repeat
Traveling Salesman Problem (TSP)
Suche Hamiltonkreis (jeder Knoten genau einmal)
(Wenn nicht möglich: A→ B → C → B → A, dann Kante ausdenken von C → A mit dem Addierten Gewicht)
Nearest Neighbor
Traveling Salesman Problem
Not optimal
Start Anywhere
Go to Nearest Unvisited City
Continue until all Cities visited
Return to Beginning
Christofides Heuristic
Create Minimum Spanning Tree
Sei M alle Kanten with odd degree Suche optimale matches zwischen allen Kanten in M, with min cost
Erstelle Eulerkreis
Erstelle Hamiltonkreis
k‐opt Heuristics
Find k edges (often 2) in a graph, so that when switching those with k different edges, the path is optimised and not broken.
Clarke and Wright Savings Algorithm
Vehicle Routing Problem
^^Called $C_{ij}$
Create savings $s_{ij}$ by $s_{ij} = c_{i0} + c_{0j} – c_{ij}$
Jede Zelle gibt an, wie viel es sparen würde direkt von Zeile → Spalte zu gehen, anstatt erst zu A zu gehen
Jetzt wird optimiert: Fangen mit dem größten Einsparungen an, und checken das der Demand $\not >$ Kapazität ist:
Maximum Flow Problem (MFP)
Weights on edges tell the max capacity. Each note should send as much as it becomes.
„Residual“-Netzwerk
The actual flow is 0, the left capacity is 5
When the actual flow increases, the the left capacity decreases by the same amount.
Augmenting Path Algorithm: Residual Capacities
Search for a Path between Start and T (regarding the capacties)
If no path exists it is optimal.
If it does fill the path to its maxima
max‐flow min‐cut theorem
A cut is defined as set of arcs so that the in and out nodes are split. Its value is the sum of all capacities.
Theorem says the smalls possible cut = maximum capacity of the flow
⇒ If a cut equals a found flow, the found flow must be optimal
Multi‐Sink Maximum Flow
Create dummy sink and sources, that connect to the real ones with arcs set to infinity.
Minimum Cost Flow Problem
Each arc has max capacity and cost.
Supply and/or demand is known.
Last changed24 days ago