Java Ant Colony Optimization Framework

jacof is an object-oriented Java-based framework for ant colony optimization (ACO). It implements the most used ACO's implementations.

Status

What is the ACO?

Dorigo [1] introduces an ant-based algorithm called Ant Colony Optimization (ACO). The algorithm tries to reproduce the behavior of the ants in the search process of solutions from the choice of the path to be followed until the process of updating the pheromone trail.

The main concepts used in ACO are based on real ants such as an artificial pheromone trail used to communication among the ants, a sequence of local moves to find the shortest path and a stochastic decision policy

The below figure shows an example of an ant building its solution traversing the graph starting from vertex 6 and reaching the vertex 4, obtaining as solution the path {6-2-3-4}.

The ACOs Developed

This framework implements the following ACO's variations:

Algorithm	Abbreviation	Authors	Year	Reference	How to use
Ant System	AS	Dorigo, Maniezzo and Colomi	1992	[1]	Click here
Elitist Ant System	EAS	Dorigo	1992	[1,3]	Click here
Ant Colony System	ACS	Dorigo and Gambardella	1997	[2]	Click here
Rank-based Ant System	ASRank	Bullnheimer, Hartl and Strauss	1997	[4]	Click here
Max-Min Ant System	MMAS	Stützle and Hoos	2000	[5]	Click here

The Problems Developed

This framework addresses the following problems:

Next Release Problem (NRP)
Symmetric Traveling Salesman Problem (TSP)
Knapsack Problem (KP)

Obs. 1: This framework supports the TSPLIB library (a library of sample instances for the TSP). Click here to access the problem instances.

Obs. 2: For Knapsack Problem, access https://people.sc.fsu.edu/~jburkardt/datasets/knapsack_01/knapsack_01.html

How To Use

To use this framework, it is necessary to define: i) the problem addressed; and ii) the ACO's variation used. After that, the algorithm should be executed. At the following it is shown some examples.

Defining the Addressed Problem

Problem nrp = new NextReleaseProblem("in/delsagrado20.nrp");
Problem tsp = new TravellingSalesmanProblem("in/oliver30.tsp");
Problem kp = new KnapsackProblem("in/p01.tsp");

Defining the ACO's variation

Some parameter settings are the same in all ACO's variations such as the number of ants, the number of iterations, alpha, betha and the evaporation rate. However, other algorithms have their own parameter settings. Next it is shown how to instanciate each ACO's variation.

Ant System

AntSystem aco = new AntSystem(problem);

aco.setNumberOfAnts(10);
aco.setNumberOfIterations(50);
aco.setAlpha(1.0);
aco.setBeta(2.0);
aco.setRho(0.1);

Ant Colony System

AntColonySystem aco = new AntColonySystem(problem);

aco.setNumberOfAnts(10);
aco.setNumberOfIterations(50);
aco.setAlpha(1.0);
aco.setBeta(2.0);
aco.setRho(0.1);
aco.setOmega(0.1);
aco.setQ0(0.9);

Elitist Ant System

ElitistAntSystem aco = new ElitistAntSystem(problem);

aco.setNumberOfAnts(10);
aco.setNumberOfIterations(50);
aco.setAlpha(1.0);
aco.setBeta(2.0);
aco.setRho(0.1);
aco.setWeight(6);

Rank-based Ant System

RankBasedAntSystem aco = new RankBasedAntSystem(problem);

aco.setNumberOfAnts(10);
aco.setNumberOfIterations(50);
aco.setAlpha(1.0);
aco.setBeta(2.0);
aco.setRho(0.1);
aco.setWeight(6);

Max-Min Ant System

MaxMinAntSystem aco = new MaxMinAntSystem(problem);

aco.setNumberOfAnts(10);
aco.setNumberOfIterations(50);
aco.setAlpha(1.0);
aco.setBeta(2.0);
aco.setRho(0.1);
aco.setStagnation(10);

Running the algorithm

To run the algorithm, you can use the ExecutionStats class defined as follows:

ExecutionStats es = ExecutionStats.execute(aco, problem);
es.printStats();

The output will be:

2017-03-31 21:01:35 [INFO ] ==================================================
2017-03-31 21:01:35 [INFO ] Execution Time: 36918.0
2017-03-31 21:01:35 [INFO ] Best Value: 423.74056313320284
2017-03-31 21:01:35 [INFO ] Best Solution: [11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 0, 1]
2017-03-31 21:01:35 [INFO ] ==================================================

References

[1] M. Dorigo. Optimization, Learning and Natural Algorithms. PhD thesis, Politecnico di Milano, Italy, 1992. [in Italian].

[2] M. Dorigo and L. M. Gambardella. Ant colony system: A cooperative learning approach to the traveling salesman problem. IEEE Transactions on Evolutionary Computation, 1(1):53–66, April 1997.

[3] M. Dorigo, V. Maniezzo, and A. Colorni. Ant system: Optimization by a colony of cooperating agents. IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, 26(1):29–41, February 1996.

[4] B. Bullnheimer, R. F. Hartl, and C. Strauss. A new rank based version of the ant system - a computational study. Central European Journal for Operations Research and Economics, 7:25–38, 1997.

[5] T. Stützle and H. H. Hoos. MAX-MIN ant system. Future Generation Computer Systems, 16(9):889–914, June 2000.

thiagodnf / jacof