Gaussian Mixture Model EM on Feature Matrix

Fits a GMM directly to a feature matrix using the EM algorithm. Unlike cluster.gmm which operates on functional data, this function works on arbitrary numeric feature matrices.

Usage

gmm.em(
  features,
  k,
  cov.type = c("full", "diagonal", "spherical"),
  max.iter = 100,
  tol = 1e-06,
  seed = 42
)

Arguments

features: Numeric matrix of features (n x d).
k: Number of mixture components.
cov.type: Covariance type: "full", "diagonal", or "spherical".
max.iter: Maximum EM iterations (default 100).
tol: Convergence tolerance (default 1e-6).
seed: Random seed.

Value

A list with components:

cluster — Integer vector of hard cluster assignments.
membership — Posterior membership probability matrix (n x k).
means — List of component mean vectors.
weights — Mixing proportions.
log.likelihood — Final log-likelihood.
bic, icl — Model selection criteria.
iterations — Number of EM iterations.

Examples

# \donttest{
X <- rbind(matrix(rnorm(100), 50, 2), matrix(rnorm(100, 3), 50, 2))
fit <- gmm.em(X, k = 2)
table(fit$cluster)
#> 
#>  2  3 
#> 50 50 
# }