The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.

Missingness-Aware Gaussian Mixture Models

Zachary McCaw

2026-02-26

Introduction

This package performs estimation and inference for Gaussian Mixture Models (GMMs) where the input data may contain missing values. Rather than imputing missing values before fitting the GMM, this package uses an extended EM algorithm to obtain the true maximum likelihood estimates of all model parameters given the observed data. In particular MGMM performs the following tasks:

Maximum likelihood estimation of cluster means, covariances, and proportions.
Calculation of cluster membership probabilities and maximum a posteriori classification of the input vectors.
Deterministic completion of the input data, by imputing missing elements to their posterior means, and stochastic completion of the input data, by drawing missing elements from the fitted GMM.

The method is detailed in Fitting Gaussian mixture models on incomplete data.

Main Functions

FitGMM estimates model parameters, performs classification and imputation.
rGMM simulates observations from a GMM, potentially with missingness.
ChooseK provides guidance on choosing the number of clusters.
GenImputation performs stochastic imputation for multiple imputation-based inference.

Compact Example

set.seed(101)
library(MGMM)

# Parameter settings.
mean_list <- list(
  c(1, 1),
  c(-1, -1)
)
cov_list <- list(
  matrix(c(1, -0.5, -0.5, 1), nrow = 2),
  matrix(c(1, 0.5, 0.5, 1), nrow = 2)
)

# Generate data.
data <- rGMM(
  n = 1e3,
  d = 2,
  k = 2,
  miss = 0.1,
  means = mean_list,
  covs = cov_list
)

# Original data.
head(data)

##           y1          y2
## 1  1.6512855  2.60621938
## 2 -0.5721069          NA
## 2 -2.0045376 -2.31888263
## 2 -0.6229388 -1.51543968
## 1  2.0258413  0.06921658
## 2 -1.3476380 -1.51915826

# Choose cluster number.
choose_k <- ChooseK(
  data,
  k0 = 2,
  k1 = 4,
  boot = 10,
  maxit = 10,
  eps = 1e-4,
  report = TRUE
)

## Cluster size 2 complete. 11 fit(s) succeeded.
## Cluster size 3 complete. 11 fit(s) succeeded.
## Cluster size 4 complete. 11 fit(s) succeeded.

# Cluster number recommendations.
show(choose_k$Choices)

##   Metric k_opt   Metric_opt k_1se   Metric_1se
## 1    BIC     2 3020.6475534     2 3020.6475534
## 2    CHI     4    4.6296476     4    4.6296476
## 3    DBI     2    0.7827537     2    0.7827537
## 4    SIL     2    0.4785446     2    0.4785446

# Estimation.
fit <- FitGMM(
  data,
  k = 2,
  maxit = 10
)

## Objective increment:  11.6 
## Objective increment:  2.72 
## Objective increment:  2.43 
## Objective increment:  2.02 
## Objective increment:  1.73 
## Objective increment:  1.49 
## Objective increment:  1.29 
## Objective increment:  1.12 
## Objective increment:  0.979 
## Objective increment:  0.86 
## 10 update(s) performed without reaching tolerance limit.

# Estimated means.
show(fit@Means)

## [[1]]
##        y1        y2 
## -1.036999 -1.052884 
## 
## [[2]]
##        y1        y2 
## 0.9515979 0.9609898

# Estimated covariances.
show(fit@Covariances)

## [[1]]
##           y1        y2
## y1 0.9647105 0.5370824
## y2 0.5370824 0.9778174
## 
## [[2]]
##            y1         y2
## y1  0.9986800 -0.4598704
## y2 -0.4598704  0.9707609

# Cluster assignments.
head(fit@Assignments)

##   Assignments      Entropy
## 1           2 9.921032e-02
## 2           1 8.258841e-01
## 2           1 5.318370e-07
## 2           1 6.179668e-03
## 1           2 9.640208e-02
## 2           1 3.354176e-04

# Deterministic imputation.
head(fit@Completed)

##           y1          y2
## 1  1.6512855  2.60621938
## 2 -0.5721069 -0.15672789
## 2 -2.0045376 -2.31888263
## 2 -0.6229388 -1.51543968
## 1  2.0258413  0.06921658
## 2 -1.3476380 -1.51915826

# Stochastic imputation.
imp <- GenImputation(fit)
head(imp)

##           y1          y2
## 1  1.6512855  2.60621938
## 2 -0.5721069  0.87181156
## 2 -2.0045376 -2.31888263
## 2 -0.6229388 -1.51543968
## 1  2.0258413  0.06921658
## 2 -1.3476380 -1.51915826

Documentation

A detailed write-up with derivations and examples is available here.

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.