Further data for the PhD thesis:

Krzysztof Socha
Ant Colony Optimization for Continuous and Mixed-Variable Optimization

Supervised by Prof. Marco Dorigo
CoDE-IRIDIA, ULB.

• Algorithms
• Test functions

Algorithms

The algorithm was implemented in R (and tested up to version 2.5.1). The following file contains the source code of the implemented algorithm: ACO.R The file contains a the main routine of both ACO_R and ACO_MV algorithms. Additional routines that make part of the algorithm are provided also in routines.R. Furthermore, file utilities.R contains some more helper functions, which are needed for flawless use of the test functions. They in particular predefine global variables if not earlier defined by the user and provide randomized rotation matrices for some of the test functions. Finally, the file nn.R provides the routines used for evaluating and training neural networks, which which were used when tackling neural network training with ACO.

In order to facilitate possible debuging of problems and issues, a test.R file is provided that illustrates the use of the algorithms and does a (high level, and not very detailed) test, if the code is producing the expected results.


Test functions

The test functions, which we used, were taken the literature (see the bibliography in the paper). The R implementation of the continuous and mixed-variable test functions may be found here: functions.R Additionally, for the evaluation of the performance of ACO on training neural networks, data were taken from the well known PROBEN1 benchmark. In particular the data related to the Heart, Diabetes, and Cancer problems were used. The R code of the test functions using this data my be respectively found in htf.R, dbf.R, and cnf.R.