Gaussian Mixture Models
GMM model a distribution using a number of gaussian distributions
dispersed across the space. The algorithm presented here uses
Expectation Maximization (EM) to fit the GMM on the data maximizing the
lower bound on likelihood.
More information on Wikipedia.
Parameters:
- # of components: number of gaussians that are used to model the distribution
- Covariance Matrix: shape of the covariance matrix used for fitting the distribution:
- Full: all elements are used
- Diagonal: only diagonal elements are used
- Spherical: identical values are used on the diagonal
- Initialization: how the means of the gaussians are initialized
- Random: randomly select a position in space
- Uniform: set the means at uniform intervals
- KMeans: use K-Means centroids as initial means