Learn R Programming

SSLfmm (version 0.1.0)

EM_FMM_SemiSupervised_Complete_Initial: Complete-Data Warm-Up Initialization for Semi-Supervised FMM with a Mixed-Missingness Mechanism

Description

Uses both labeled and unlabeled subsets of the data to obtain quick initial estimates for mixture parameters and missingness mechanism parameters (alpha, xi) via a warm-up EM procedure.

Usage

EM_FMM_SemiSupervised_Complete_Initial(
  data,
  g = 2,
  ncov = 1,
  alpha_init = 0.01,
  warm_up_iter = 200,
  tol = 1e-06
)

Value

A list with initial values for EM_FMM_SemiSupervised:

  • pi - mixture weights.

  • mu - list of component mean vectors.

  • Sigma - covariance matrix/matrices.

  • alpha - MCAR proportion.

  • xi - logistic regression coefficients for MAR mechanism.

Arguments

data

A data frame containing:

  • The first p columns: numeric variables used in the FMM.

  • A column missing: indicator (0 = labeled, 1 = unlabeled/missing).

  • A column z: class labels for labeled rows (1:g); NA for unlabeled.

g

Integer, number of mixture components (default 2).

ncov

Integer, covariance structure: 1 = shared (equal), 2 = class-specific (unequal).

alpha_init

Numeric in (0,1), initial MCAR proportion (default 0.01).

warm_up_iter

Integer, number of warm-up EM iterations (default 200).

tol

Convergence tolerance on alpha (default 1e-6).

Details

  • This function first calls initialestimate to get initial \(\pi\), \(\mu\), \(\Sigma\).

  • Then it calls EM_FMM_SemiSupervised_Initial with these values for a short warm-up run.

  • Covariance structure (equal vs. unequal) is determined by ncov.