Unsupervised Clustering with Gaussian Mixture Modeling

Summary

This repository contains code for performing Gaussian mixture modeling (GMM) to separate two-dimensional datasets into classes by modeling the data as samples from two or more Gaussian distributions.

How to Use

A full description of function inputs/outputs can be viewed by using help smart.cluster.gaussian in the MATLAB command line. Generally, the function's inputs are the unlabeled data and model hyperparameters, and the output is a struct containing class labels for each input datapoint and the final model parameters contained in a struct. This output can be used to seed the model on subsequent function calls. The following is an example function call:

new_model = smart.cluster.gaussian(data, true, 'startModel', old_model, 'numClusters', 3, 'iterations', 100);
labeled_data = new_model.indices; old_model = new_model.model;

Note that the +cluster package is contained within the +smart package for compatibility with other repositories, since they may contain other subpackages of the larger +smart package which rely on +cluster package functions. These other subpackages may be pasted in the +smart folder to add their respective functionality, though they will generally already contain all dependencies when cloned.

Data Formatting

Data should be provided in a Mx2 matrix with M observations of two-dimensional data.

Member Functions

Function	Purpose
`evaluate()`	Compute the silhouette score for clustered data
`gaussian()`	Perform Gaussian mixture modeling

References