Skip to contents

Gaussian Mixture Model Clustering for Functional Data

Source: R/clustering.R
cluster.gmm.Rd

Performs model-based clustering using Gaussian mixture models on functional data. Curves are projected onto a B-spline or Fourier basis before fitting. Automatic K selection via BIC or ICL.

Usage

cluster.gmm(
  fdataobj,
  k.range = 2:6,
  covariates = NULL,
  nbasis = 10,
  basis.type = c("bspline", "fourier"),
  cov.type = c("full", "diagonal"),
  cov.weight = 1,
  max.iter = 100,
  tol = 1e-04,
  n.init = 5,
  seed = NULL,
  criterion = c("bic", "icl")
)

Arguments

fdataobj

An object of class 'fdata'.

k.range

Range of number of clusters to try (default 2:6).

covariates

Optional matrix of scalar covariates to include.

nbasis

Number of basis functions for projection (default 10).

basis.type

Basis type: "bspline" (default) or "fourier".

cov.type

Covariance structure: "full" (default) or "diagonal".

cov.weight

Weight for covariates vs basis coefficients (default 1).

max.iter

Maximum EM iterations (default 100).

tol

Convergence tolerance (default 1e-4).

n.init

Number of random initializations (default 5).

seed

Optional random seed.

criterion

Model selection criterion: "bic" (default) or "icl".

Value

An object of class 'cluster.gmm' with components:

cluster

Integer vector of cluster assignments (1-indexed)

membership

Matrix of posterior membership probabilities

means

Component means matrix

weights

Mixing proportions

bic

BIC of selected model

icl

ICL of selected model

k

Number of clusters selected

converged

Whether EM converged

bic.values

BIC values for each K tried

See also

cluster.kmeans for hard clustering, cluster.fcm for fuzzy clustering

Examples

# \donttest{
fd <- fdata(rbind(matrix(rnorm(250), 25, 10),
                  matrix(rnorm(250, 2), 25, 10)),
            argvals = seq(0, 1, length.out = 10))
gmm <- cluster.gmm(fd, k.range = 2:3)
# }