Package 'mixturemodelsr'

Compute AIC for fitted model

Description

Compute AIC for fitted model

Usage

mm_aic(fit)

Arguments

fit	An mm_fit object

Value

Numeric AIC value

Examples

## Not run: 
fit <- mm_gmm_fit(iris[,1:4], k = 3)
mm_aic(fit)

## End(Not run)

---

Compute BIC for fitted model

Description

Compute BIC for fitted model

Usage

mm_bic(fit)

Arguments

fit	An mm_fit object

Value

Numeric BIC value

Examples

## Not run: 
fit <- mm_gmm_fit(iris[,1:4], k = 3)
mm_bic(fit)

## End(Not run)

---

Fit a Constrained Gaussian Mixture Model

Description

Fits a GMM with a common covariance matrix across all components.

Usage

mm_gmm_constrained_fit(
  x,
  k,
  optimizer = "Newton-CG",
  scale = 1,
  use_kmeans = TRUE,
  ...
)

Arguments

x	Numeric matrix or data.frame (rows = observations, columns = features)
k	Number of mixture components
optimizer	Optimizer name: "Newton-CG" (default), "grad_descent", "rms_prop", or "adam"
scale	Initialization scale parameter (default 1.0)
use_kmeans	Logical, whether to use k-means for initializing component means (default TRUE)
...	Additional arguments passed to Python init_params() or fit() methods

Value

An mm_fit object

Examples

## Not run: 
fit <- mm_gmm_constrained_fit(iris[, 1:4], k = 3)
mm_bic(fit)

## End(Not run)

---

Fit a Gaussian Mixture Model (GMM)

Description

Fits a Gaussian mixture model to data using gradient-based optimization. This is a wrapper around the Python Mixture-Models GMM implementation.

Usage

mm_gmm_fit(x, k, optimizer = "Newton-CG", scale = 1, use_kmeans = TRUE, ...)

Arguments

x	Numeric matrix or data.frame (rows = observations, columns = features)
k	Number of mixture components
optimizer	Optimizer name: "Newton-CG" (default), "grad_descent", "rms_prop", or "adam"
scale	Initialization scale parameter (default 1.0)
use_kmeans	Logical, whether to use k-means for initializing component means (default TRUE)
...	Additional arguments passed to Python init_params() or fit() methods

Value

An mm_fit object containing:

py_model	Python model object
params_store	Full optimization path
params	Final fitted parameters
model_name	Model family name
k	Number of components
call	Original function call
n	Sample size
d	Number of dimensions
optimizer	Optimizer used

Examples

## Not run: 
# Setup (first time only)
mm_setup()

# Fit a 3-component GMM on iris data
fit <- mm_gmm_fit(iris[, 1:4], k = 3)
print(fit)

# Get cluster labels
labels <- mm_predict(fit)

# Model selection
mm_bic(fit)
mm_aic(fit)

## End(Not run)

---

Compute log-likelihood for fitted model

Description

Compute log-likelihood for fitted model

Usage

mm_likelihood(fit)

Arguments

fit	An mm_fit object

Value

Numeric log-likelihood value

Examples

## Not run: 
fit <- mm_gmm_fit(iris[,1:4], k = 3)
mm_likelihood(fit)

## End(Not run)

---

Fit an MCLUST Family Model

Description

Fits a model from the MCLUST family of constrained Gaussian mixture models. MCLUST models specify different parameterizations of the covariance structure.

Usage

mm_mclust_fit(
  x,
  k,
  model_type = NULL,
  optimizer = "Newton-CG",
  scale = 1,
  use_kmeans = TRUE,
  ...
)

Arguments

x	Numeric matrix or data.frame (rows = observations, columns = features)
k	Number of mixture components
model_type	Character string specifying the MCLUST model type. Common types include: "EII", "VII", "EEI", "VEI", "EVI", "VVI", "EEE", "VEE", "EVE", "VVE", "EEV", "VEV", "EVV", "VVV" If NULL, uses default from Python implementation.
optimizer	Optimizer name: "Newton-CG" (default), "grad_descent", "rms_prop", or "adam"
scale	Initialization scale parameter (default 1.0)
use_kmeans	Logical, whether to use k-means for initializing component means (default TRUE)
...	Additional arguments passed to Python init_params() or fit() methods

Details

MCLUST model types follow a three-letter naming convention:

• First letter: Volume (E=equal across components, V=variable across components)

• Second letter: Shape (E=equal, V=variable, I=spherical/identity)

• Third letter: Orientation (E=equal, V=variable, I=axis-aligned/identity)

Common model types:

• "EII": Spherical, equal volume

• "VII": Spherical, variable volume

• "EEE": Ellipsoidal, equal volume, shape, and orientation

• "VVV": Ellipsoidal, variable volume, shape, and orientation (most flexible)

Value

An mm_fit object

Examples

## Not run: 
# Fit MCLUST with VVV (most flexible) model
fit <- mm_mclust_fit(iris[, 1:4], k = 3, model_type = "VVV")
mm_bic(fit)

# Fit MCLUST with spherical, equal volume model
fit_eii <- mm_mclust_fit(iris[, 1:4], k = 3, model_type = "EII")
mm_bic(fit_eii)

# Compare models
labels <- mm_predict(fit)
table(labels, iris$Species)

## End(Not run)

---

Fit a Mixture of Factor Analyzers (MFA)

Description

Fits a mixture of factor analyzers model to data using gradient-based optimization.

Usage

mm_mfa_fit(
  x,
  k,
  q = NULL,
  optimizer = "Newton-CG",
  scale = 1,
  use_kmeans = TRUE,
  ...
)

Arguments

x	Numeric matrix or data.frame (rows = observations, columns = features)
k	Number of mixture components
q	Number of latent factors (NULL for automatic selection)
optimizer	Optimizer name: "Newton-CG" (default), "grad_descent", "rms_prop", or "adam"
scale	Initialization scale parameter (default 1.0)
use_kmeans	Logical, whether to use k-means for initializing component means (default TRUE)
...	Additional arguments passed to Python init_params() or fit() methods

Value

An mm_fit object

Examples

## Not run: 
# Fit MFA with 3 components and 2 latent factors
fit <- mm_mfa_fit(iris[, 1:4], k = 3, q = 2)
mm_bic(fit)

# Get cluster labels
labels <- mm_predict(fit)

## End(Not run)

---

Get parameter values from fit object

Description

Extracts the parameter values from a fitted mixture model.

Usage

mm_params(fit, convert = TRUE)

Arguments

fit	An mm_fit object
convert	Logical, whether to convert Python objects to R (default TRUE)

Value

Parameter object (structure depends on model family)

Examples

## Not run: 
fit <- mm_gmm_fit(iris[,1:4], k = 3)
params <- mm_params(fit)

## End(Not run)

---

Fit a Parsimonious Gaussian Mixture Model (PGMM)

Description

Fits a parsimonious GMM with constraints on the covariance structure.

Usage

mm_pgmm_fit(
  x,
  k,
  model_type = NULL,
  q = NULL,
  optimizer = "Newton-CG",
  scale = 1,
  use_kmeans = TRUE,
  ...
)

Arguments

x	Numeric matrix or data.frame (rows = observations, columns = features)
k	Number of mixture components
model_type	Character string specifying the PGMM model type (e.g., "VVV", "EEE", "VEV"). If NULL, uses default from Python implementation.
q	Number of latent factors (NULL for automatic selection)
optimizer	Optimizer name: "Newton-CG" (default), "grad_descent", "rms_prop", or "adam"
scale	Initialization scale parameter (default 1.0)
use_kmeans	Logical, whether to use k-means for initializing component means (default TRUE)
...	Additional arguments passed to Python init_params() or fit() methods

Details

PGMM model types follow the mclust naming convention:

• First letter: Volume (E=equal, V=variable)

• Second letter: Shape (E=equal, V=variable)

• Third letter: Orientation (E=equal, V=variable, I=identity)

Value

An mm_fit object

Examples

## Not run: 
# Fit PGMM with variable volume, shape, and orientation
fit <- mm_pgmm_fit(iris[, 1:4], k = 3, model_type = "VVV")
mm_bic(fit)

## End(Not run)

---

Predict cluster labels

Description

Predicts cluster membership labels for observations using a fitted mixture model.

Usage

mm_predict(fit, newx = NULL)

Arguments

fit	An mm_fit object returned by mm_*_fit functions
newx	Optional matrix or data.frame of new observations. If NULL, predictions are made on the training data.

Value

Integer vector of cluster labels (0-indexed from Python, converted to 1-indexed for R)

Examples

## Not run: 
fit <- mm_gmm_fit(iris[,1:4], k = 3)
labels <- mm_predict(fit)
table(labels, iris$Species)

## End(Not run)

---

Python Configuration Diagnostics

Description

Prints detailed information about the Python environment configuration for mixturemodelsr. Useful for troubleshooting installation issues.

Usage

mm_py_info()

Value

Invisibly returns a list with diagnostic information

Examples

## Not run: 
mm_py_info()

## End(Not run)

---

Check if Python module is available

Description

Check if Python module is available

Usage

mm_python_available()

Value

Logical indicating if either mixture_models or Mixture_Models is importable

---

Setup mixturemodelsr (user-friendly wrapper)

Description

One-time setup for R users. Provisions a dedicated conda environment with Python 3.10 + NumPy 1.23.5 and installs Mixture-Models==0.0.8.

Usage

mm_setup(force = FALSE)

Arguments

force

Logical, force reinstallation by deleting and recreating the env

Value

TRUE invisibly on success

---

Fit a t-Mixture Model (TMM)

Description

Fits a mixture of multivariate t-distributions to data using gradient-based optimization. T-distributions are more robust to outliers than Gaussian distributions.

Usage

mm_tmm_fit(x, k, optimizer = "Newton-CG", scale = 1, use_kmeans = TRUE, ...)

Arguments

x	Numeric matrix or data.frame (rows = observations, columns = features)
k	Number of mixture components
optimizer	Optimizer name: "Newton-CG" (default), "grad_descent", "rms_prop", or "adam"
scale	Initialization scale parameter (default 1.0)
use_kmeans	Logical, whether to use k-means for initializing component means (default TRUE)
...	Additional arguments passed to Python init_params() or fit() methods

Details

T-mixture models use multivariate t-distributions instead of Gaussians, making them more robust to outliers. Each component has its own degrees of freedom parameter, allowing different tail behaviors.

Value

An mm_fit object

Examples

## Not run: 
# Fit TMM with 3 components (robust to outliers)
fit <- mm_tmm_fit(iris[, 1:4], k = 3)
mm_bic(fit)

# Get cluster labels
labels <- mm_predict(fit)
table(labels, iris$Species)

## End(Not run)

---

Print method for mm_fit objects

Description

Print method for mm_fit objects

Usage

## S3 method for class 'mm_fit'
print(x, ...)

Arguments

x	An mm_fit object
...	Additional arguments (ignored)

---