ABC fuzzy clustering

The goal of this project is to create a fuzzy clustering algorithm based on the artificial bee colony algorithm. Robustness to the problems of the fuzzy c-means algorithm, in particular when handling many-dimensional data, is important.

Compilation

To compile the C++ code as a Python extension, see the official guide for Unix-like systems and Windows.

The VECTOR_DIM preprocessor macro defines the dimensionality of the vectors to be clustered. Wrong dimensionality will result in a runtime error. To cluster vectors of different size than the default 2, override the macro using compiler parameters or manually overwrite it in abc_plusplus.cpp.

Python API

The abc_plusplus module contains 4 classes:

• ArtificialBeeColony - unmodified ABC
• ModArtificialBeeColony - ABC with DE-inspired mixing stragey
• TournamentArtificialBeeColony - ABC with tournament selection startegy
• TournamentModArtificialBeeColony - ABC with both modifications

The constructors for ArtificialBeeColony and TournamentArtificialBeeColony accept 4 positional parameters:

• the data to be clustered, represented as an n by m sequence of floats, where n is the number of vectors and m is dimensionality
• the number of clusters (a positive integer)
• the size of the population (a positive integer)
• the maximum number of iterations for which a solution is retained without any improvement (a positive integer)

The constructors for ModArtificialBeeColony and TournamentModArtificialBeeColony accept 2 additional positional parameters, 6 in total:

• the scale factor (a float between 0 and 1)
• the modification rate (a float between 0 and 1)

All classes define 3 methods:

• optimize - takes one parameter - the number of iterations. Runs the algorithm for the specified number of iterations and returns the best found solution (as a list of lists of floats). The population is retained between calls.
• fit - similar to optimize, but without a return value.
• score - returns the fitness score (as a float) of the best currently know solution.

Acknowledgements

Basic algorithm based on:

Karaboga, D., Ozturk, C.: Fuzzy clustering with artificial bee colony algorithm.Scientific Research and Essays5(07 2010)

Modifications based on:

Kumar, A., Kumar, D., Jarial, S.: A hybrid clustering method based on improvedartificial bee colony and fuzzy c-means algorithm. International Journal of ArtificialIntelligence15, 40–60 (01 2017)

and:

Ouadfel, S., Meshoul, S.: Handling fuzzy image clustering with a modified abcalgorithm. International Journal of Intelligent Systems and Applications4, 65–74(11 2012).

The DIM-set dataset is from:

P. Fränti, O. Virmajoki and V. Hautamäki, "Fast agglomerative clustering using a k-nearest neighbor graph", IEEE Trans. on Pattern Analysis and Machine Intelligence, 28 (11), 1875-1881, November 2006.

The Worms dataset is from:

S. Sieranoja and P. Fränti, "Fast and general density peaks clustering", Pattern Recognition Letters, 128, 551-558, December 2019.