Provides rough initial estimates of the missingness parameters alpha
and xi, together with mixture parameters pi, mu, and
covariance matrices, using a lightweight EM-style routine. The covariance
structure is chosen automatically based on Sigma_init:
If Sigma_init is a \(p \times p\) matrix, a
shared (equal) covariance is used.
If Sigma_init is a list of length g of
\(p \times p\) matrices or a \(p \times p \times g\) array,
class-specific (unequal) covariances are used.
If Sigma_init is NULL, a shared covariance is
estimated from the labeled data.
This function is intended as a fast, heuristic initializer rather than a final estimator for the mixed missingness model.
EM_FMM_SemiSupervised_Initial(
Y_labelled,
Z_labelled,
Y_unlabelled,
g = 2,
pi_init = NULL,
mu_init = NULL,
Sigma_init = NULL,
alpha_init = 0.01,
warm_up_iter = 50,
tol = 1e-06
)A list with elements:
pi - length-g vector of mixing proportions.
mu - list of g mean vectors.
Sigma - shared \(p \times p\) matrix (equal-Sigma) or list
of g matrices (unequal-Sigma).
xi - length-2 numeric vector c(xi0, xi1) from the
logistic MAR model.
alpha - estimated MCAR proportion.
gamma - \(n \times g\) responsibility matrix.
d2_yj - numeric vector of entropy-based scores used in the
missingness model.
Numeric matrix of labeled observations (\(n_L \times p\)).
Integer vector of class labels in 1:g for
Y_labelled.
Numeric matrix of unlabeled observations (\(n_U \times p\)).
Integer, number of mixture components (default 2).
Optional numeric length-g vector of initial mixing
proportions.
Optional list of length g of initial mean vectors
(each of length p).
Optional initial covariance:
a \(p \times p\) matrix (shared), or a list of g
\(p \times p\) matrices, or a \(p \times p \times g\) array
(class-specific).
Numeric in \((0,1)\), initial MCAR proportion
(default 0.01).
Integer, number of warm-up EM iterations used to
refine the quick initial estimates (default 50).
Convergence tolerance on alpha (default 1e-6).