UnsupervisedClustering.jl

A Julia package providing a unified interface for unsupervised clustering algorithms with advanced optimization techniques to escape local optima and reduce overfitting.

Key Features

• Unified Interface: Almost all clustering algorithms use the same fit(algorithm, data, k) pattern (the kmedoids algorithm is an exception, requiring a distance matrix)
• Advanced Optimization: Metaheuristic approaches including genetic algorithms and multi-start strategies
• Type-Stable: Modern Julia design with parameterized types for zero runtime overhead
• Composable: Mix and match algorithms through chaining and metaheuristic composition

Quick Start

using UnsupervisedClustering
using RegularizedCovarianceMatrices

# Generate sample data
n = 100
d = 2
k = 3
data = rand(n, d)

# Local search algorithms
kmeans = Kmeans()
gmm = GMM(estimator = EmpiricalCovarianceMatrix(n, d))

# Metaheuristic algorithms
genetic = GeneticAlgorithm(local_search = kmeans)
multi_start = MultiStart(local_search = kmeans)
random_swap = RandomSwap(local_search = kmeans)

# All use the same fit function!
result1 = fit(kmeans, data, k)
result2 = fit(genetic, data, k)
result3 = fit(multi_start, data, k)

# The unified interface enables some compositions:
chain = ClusteringChain(kmeans, gmm)
result4 = fit(chain, data, k)

Algorithm Categories

Local Search Algorithms

• K-means: Classic centroid-based clustering with Lloyd's algorithm
• K-means++: Improved initialization for better clustering quality
• K-medoids: Robust clustering using actual data points as centers
• GMM: Gaussian Mixture Models with EM algorithm

Metaheuristic Algorithms

• Genetic Algorithm: Evolutionary approach for global optimization
• Multi-Start: Multiple random initializations with best result selection
• Random Swap: Perturbation-based local search escape

Ensemble Methods

• Clustering Chain: Sequential algorithm composition for refined results

Installation

using Pkg
Pkg.add("UnsupervisedClustering")

Related Packages

This package integrates with:

• RegularizedCovarianceMatrices.jl: For advanced GMM regularization techniques
• Distances.jl: For distance metrics in clustering algorithms

Citing

If you find UnsupervisedClustering useful in your work, we kindly request that you cite the following paper:

@article{sampaio2024regularization,
  title={Regularization and optimization in model-based clustering},
  author={Sampaio, Raphael Araujo and Garcia, Joaquim Dias and Poggi, Marcus and Vidal, Thibaut},
  journal={Pattern Recognition},
  pages={110310},
  year={2024},
  publisher={Elsevier}
}