devasena37 / psoclustering

This is an implementation of clustering IRIS dataset with particle swarm optimization(PSO)

PSOClustering

This is an implementation of clustering data with particle swarm optimization(PSO) algorithm

This implementation is inspired by the following paper : Data Clustering using Particle Swarm Optimization

The dataset used in this implementation is the IRIS flower dataset but this can surely work with other datasets too!

Run

after installing the requirements run main.py. you can choose to plot the point with matplotlib by changing the variable plot at the top of the code.

However be aware that this will only keep the first 2 dimensions of the dataset points and other dimensions will not be considered.

Other than the number of clusters (n_clusters) and the number of particles (n_particles) you can choose whether to use kmeans for seeding the initial swarm( called Hybrid PSO Clustering) or not ( with hybrid variable). You can also change the pso algorithm parameters w, c1 and c2.

Usage

will need pso_clustering.py and particle.py.

from pso_clustering import PSOClusteringSwarm

# data should be a numpy array of n-dimensional points

pso = PSOClusteringSwarm(n_clusters=3, n_particles=10, data=data_points, hybrid=True, w=0.72, c1=1.49, c2=1.49)

clusters, global_best_fitness = pso.start(iteration=1000)

The function start() will return a tuple of the final clusters (for each data point has the cluster id) and the final value of the global best fitness of the swarm.