Solution methodologies for the classical assignment problem Grant van Dieman Friday 30th November 2007 Supervisor: Prof. JH van Vuuren Co-supervisor: Mr JN Roux Overview The classical assignment problem Exact Solution methods A maximum matching algorithm Successive shortest path method Hungarian method Greedy heuristics Comparison Future work Slide 2 The classical assignment problem Votaw and Orden (1952) Assumptions xij is 1 if assignee i is assigned to task j and 0 otherwise The assignment problem is NP complete (Lloyd and Witzenhausen (1986)) Slide 3 The Weapon Target Assignment Problem Flood (1957) Vj : priority of eliminating target j. qij : is the survival probability of target j if it is engaged by weapon i. xij =1 if weapon i engage target j and 0 otherwise Slide 4 Overview The classical assignment problem Exact Solution methods A maximum matching algorithm Successive shortest path method Hungarian method Greedy heuristics Comparison Future work Slide 5 A maximum matching algorithm for weighted bipartite graphs (MWM) V1 = {assignees} G: qij V2 = {tasks} Slide 6 A maximum matching algorithm for weighted bipartite graphs (MWM) V1 = {assignees} M: qij V2 = {tasks} Slide 7 Overview The classical assignment problem Exact Solution methods A maximum matching algorithm Successive shortest path method Hungarian method Greedy heuristics Comparison Future work Slide 8 Successive shortest path algorithm (SSP) Minimum cost flow algorithm Minimize cij xij i,j E subject to x j: j ,i E ji x j: i , j E ij bi xij uij xij lij i V , i, j E , i, j E. Why this algorithm can be used to solve the assignment problem The value of xij will be binary Slide 9 Overview The classical assignment problem Exact Solution methods Successive shortest path method A maximum matching algorithm Hungarian method Greedy heuristics Comparison Future work Slide 10 Hungarian Method Kuhn(1955) Special algorithm for the assignment problem Construct reduced cost matrix Slide 11 Overview The classical assignment problem Exact Solution methods Successive shortest path method A maximum matching algorithm Hungarian method Greedy heuristics Comparison Future work Slide 12 Greedy Heuristics Greedy RTB Greedy RBT Greedy RR Greedy CLR Greedy CRL Greedy CR Slide 13 Overview The classical assignment problem Exact Solution methods Successive shortest path method A maximum matching algorithm Hungarian method Greedy heuristics Comparison Future work Slide 14 Comparisons Solution times of exact methods Benchmark set 1: JE Beasly (Randomly Generated) 60.0 1024 MB ram, Windows XP 50 3.4 Ghz, time (seconds) 40 35 30 40.0 30.0 25 20 20.0 15 10 10.0 5 0 0.0 number of iterations 45 50.0 100 200 300 400 500 600 700 800 Hungarian Hungarian 0.5 0.4 1.0 1.9 3.1 3.9 5.1 6.1 MWM MWM 1.6 8.9 10.9 37.5 36.4 9.9 MWM Iterations 44 38 14 15 7 size Slide 15 2 49.3 23.4 4 3 MWM Iterations Expon. (MWM) Comparisons Solution times 6 RTB time (seconds) 5 RBT RR 4 CLR CRL 3 CR 2 1 0 100 200 300 400 500 size Slide 16 600 700 800 Comparisons % away from optimal 1.2 % optimal 1 RTB RBT 0.8 RR 0.6 CLR CRL 0.4 CR 0.2 0 100 200 300 400 500 size Slide 17 600 700 800 Comparisons Solution times of exact methods Benchmarks set 2: Randomly Generated in Matlab 1200 800 600 Hungarian 400 MWM 200 SSP 60 70 80 90 10 0 20 0 30 0 40 0 50 0 60 0 70 0 80 0 90 10 0 00 0 10 20 30 40 50 time (seconds) 1000 size Slide 18 Expon. (SSP) Comparisons Solution time 300 RTB 200 RBT RR 150 CLR CRL 100 CR 50 size Slide 19 80 0 10 00 30 00 60 0 40 0 20 0 90 70 50 30 0 10 time (seconds) 250 Comparisons % away from optimal 4 3.5 RTB RBT 2.5 RR 2 CLR 1.5 CRL CR 1 0.5 size Slide 20 30 00 10 00 80 0 60 0 40 0 20 0 90 70 50 30 0 10 % optimal 3 Future work Advanced Heuristics and Meta-heuristics More exact solution methods Expand algorithms to solve variations of the assignment problem Slide 21 References [1] [2] [3] [4] [5] Slide 22