One-Hot Encodes Factor Variables (FIRST Level as Reference)

Description

This function takes a dataframe, identifies all columns of class factor, and converts them into **dummy variables** using one-hot encoding via stats::model.matrix. For each factor, the function explicitly removes the first dummy variable generated, effectively making the **first level** of the factor the **reference level** (omitted category). Non-factor columns are retained as is.

Usage

encodeCat(dataframe)

Arguments

Value

A data.frame where:

• All original non-factor columns are present.

• All original factor columns are replaced by a set of binary (0/1) dummy variables. The first level of the factor is excluded from the generated dummies, making the last level the reference.

Examples

data("exposure_data")
exp_data = exposure_data$df
covList = {}
covList$FE = c('X')
XFE = encodeCat(exp_data[,covList$FE, drop = FALSE])

List of the different outputs of the main function for examples

Description

A list of the different outputs of the main function for examples

Usage

examp

Format

A list with 4 components:

dataProfile

Output of the profileGLMM_preprocess() function example

MCMC_Obj

Output of the profileGLMM_Gibbs() function example

post_Obj

Output of the profileGLMM_postprocess() function example

pred_Obj

Output of the profileGLMM_predict() function example

Source

Generated synthetically by the package authors.

Simulated Data and Parameters for a exposure profile linear mixed model

Description

A list containing a simulated exposure dataset (df) and the ground-truth parameters (theta0) used to generate it.

The dataset df contains N = 4500 observations across n_{Ind} = 1500 individuals, with $n_R = 3$ repeated measures per individual.

Usage

exposure_data

Format

A list with 2 components:

df

A data frame with 4,500 rows and 6 variables (the simulated data).

theta0

A list of 11 elements containing the true parameters used for simulation.

Details

The underlying model for the response \bold{Y} is:

\bold{Y} = \bold{X}_{Fe}\bold{\beta} + \bold{X}_{Int}\bold{\alpha}_{Lat} + \bold{X}_{Re}\bold{\alpha}_{RE} + \bold{\epsilon}

df Data Variables

X

Continuous predictor (\sim N(0, 1)).

t

Time-like variable (structured around 0, 1, 2).

indiv

**Individual ID** (1 to 1500), the grouping factor.

Exp1, Exp2

Exposure continuous predictors.

Y

The **Simulated Response Variable** calculated as: \bold{Y} = y_{Fe} + y_{Int} + y_{Re} + \epsilon, where \epsilon ~ N(0, 1).

theta0 Parameters

The list theta0 holds the true values used to generate Y, including:

• Lat: **Categorical Factor** (9 levels), defining the clusters for interaction effects.

• beta: True fixed effects for the global intercept and \bold{X} (i.e., $(3, 2)$).

• alphaLat: Vector of 18 coefficients defining the cluster-specific intercepts and slopes for \bold{X} within the 9 Lat categories.

• alphaRE: Vector of 1500 random slopes for the time variable \bold{t}, drawn from $N(0, 1)$.

• sigma: Residual standard deviation (1).

Source

Generated synthetically by the package authors.

Simulated Data and Parameters for a Piecewise Example

Description

A list containing a second simulated dataset (df) and its ground-truth parameters (theta0). This dataset is generated from a **piecewise linear model**, where the continuous predictor x is segmented into 6 bins, and different intercept and slope coefficients are applied to each segment.

The dataset df contains $N = 3000$ observations.

Usage

piecewise_data

Format

A list with 2 components:

df

A data frame with 3,000 rows and 2 variables (the simulated data).

theta0

A list of 5 elements containing the true parameters used for simulation.

Details

The underlying model for the response \bold{Y} is:

\bold{Y} = \bold{X}_{Fe}\bold{\beta} + \bold{X}_{Lat}\bold{\alpha}_{Lat} + \bold{\epsilon}

where \bold{X}_{Fe} is the global intercept, and \bold{X}_{Lat}\bold{\alpha}_{Lat} models the piecewise relationship of x across the 6 categories defined in theta0$Lat. The error term \bold{\epsilon} ~ N(0, 1).

df Data Variables

x

A continuous predictor, uniformly distributed between -3 and 3.

Y

The **Simulated Response Variable** defined by the piecewise linear model.

theta0 Parameters

The list theta0 holds the true values used for simulation, including:

• beta: True global intercept (i.e., (0.5)).

• Lat: The categorical factor (1 to 6) derived from segmenting x.

• alphaLat: Vector of $2 * 6 = 12$ coefficients defining the specific intercept and slope for x within each of the 6 segments.

Source

Generated synthetically by the package authors.

Initialize the prior hyperparameters for the Profile GLMM

Description

This function establishes the prior distributions for all parameters in the Profile GLMM. It sets up vague, non-informative priors (often using small precision/large variance or conjugate forms like Wishart/Dirichlet) for the fixed effects (beta_{FE}), residual variance (\sigma^2), random effects covariance (\Sigma_{RE}), latent effects covariance (\Sigma_{Lat}), cluster parameters (means and covariances), and the Dirichlet Process parameters (\alpha).

Usage

prior_init(params)

Arguments

Value

A list (prior) containing the hyperparameter values structured by the parameter block they govern:

FE:

Priors for fixed effects and residual variance (e.g., lambda, a, b for conjugate Normal-Gamma).

RE:

Inverse-Wishart priors for random effects covariance (\Sigma_{RE}) (e.g., Phi, eta).

assign:

Priors for the cluster assignment parameters, nested under Cont (Normal-Inverse-Wishart for continuous) and Cat (Dirichlet for categorical).

Lat:

Inverse-Wishart prior for the latent effects covariance (\Sigma_{Lat}) (e.g., Phi, eta).

DP:

Parameters for the Dirichlet Process prior (e.g., scale, shape).

Examples

# Load dataProfile, the result of profileGLMM_preProcess()
data("examp")
dataProfile = examp$dataProfile
prior_config <- prior_init(dataProfile$params)

R Wrapper for Profile GLMM Gibbs Sampler (C++ backend)

Description

This is the main function for fitting the Profile Generalized Linear Mixed Model (Profile GLMM) using a blocked Gibbs sampling algorithm. It acts as an R wrapper, passing pre-processed data, initial values, and prior hyperparameters contained in the model object directly to the C++ implementation GSLoopCPP. The function simulates the posterior distribution of all model parameters, including fixed effects, random effects variance, profile cluster parameters, latent effects, and cluster assignments.

Usage

profileGLMM_Gibbs(model, nIt, nBurnIn)

Arguments

Value

A list containing the saved Gibbs-sampled MCMC chains for all model parameters (e.g., beta, Z, gamma, pvec, muClus, PhiClus, etc.) and the variable names from the original data. This output is ready for post-processing with profileGLMM_postProcess.

Examples

# Load dataProfile, the result of profileGLMM_pREProcess()
data("examp")
dataProfile = examp$dataProfile
MCMC_Obj = profileGLMM_Gibbs(model = dataProfile,
   nIt = 100,
   nBurnIn = 10
 )

Post-process the MCMC chain from profileGLMM_Gibbs

Description

This function performs essential post-processing of the MCMC output generated by the profileGLMM_Gibbs function. It calculates the posterior means and credible intervals for the fixed effects (population parameters) and, optionally, computes a representative cluster partition using methods like Least Squares (LS) or Ng's spectral clustering (NG) on the co-occurrence matrix. It also provides estimated cluster characteristics (centroids, probability vectors, and outcome effects) for the representative partition.

Usage

profileGLMM_postProcess(
  MCMC_Obj,
  modeClus = "NG",
  comp_cooc = TRUE,
  alpha = 0.05
)

Arguments

Value

A list with three elements:

coocMat:

The co-occurrence matrix of the MCMC cluster assignments (MCMC_Obj$Z).

clust:

A list containing the results of the representative clustering (if comp_cooc = TRUE), including the optimal partition (Zstar), number of clusters (Kstar), representative cluster parameters (cen, pvec, gamma), and full posterior samples for the cluster characteristics.

pop:

A list containing the posterior mean and (1-alpha) credible intervals for the fixed effects (betaFE).

Examples

# Load MCMC_Obj, the result of profileGLMM_Gibbs()
data("examp")
MCMC_Obj = examp$MCMC_Obj
post_Obj = profileGLMM_postProcess(MCMC_Obj, modeClus='LS')
print(post_Obj$pop$betaFE)

Prediction of cluster memberships and outcomes

Description

This function uses the results of the post-processed Profile GLMM MCMC chain to predict cluster memberships and outcomes for new or existing data. It first calculates the fixed effect (FE) contribution and then, if a representative clustering is available in post_Obj, computes the predicted cluster membership and the corresponding latent effect (Lat) contribution to the outcome.

Usage

profileGLMM_predict(post_Obj, XFE, XLat, UCont, UCat)

Arguments

Value

A list with the following elements:

FE:

A numeric vector of the predicted fixed effects contribution to the outcome.

Y:

A numeric vector of the total predicted outcome (FE + Lat).

classPred:

A factor vector of the predicted cluster membership for each observation. NULL if no representative clustering was provided in post_Obj.

Int:

A numeric vector of the predicted latent effect contribution to the outcome. NULL if no representative clustering was provided.

Examples

# Load post_Obj, the result of profileGLMM_postProcess()
data("examp")
post_Obj = examp$post_Obj
# Load dataProfile, the result of profileGLMM_preProcess()
dataProfile = examp$dataProfile
pred_Obj = profileGLMM_predict(post_Obj,
                               dataProfile$d$XFE,
                               dataProfile$d$XLat,
                               dataProfile$d$UCont,
                               dataProfile$d$UCat)

Preprocess the data from a list describing the profile LMM model

Description

Preprocess the data from a list describing the profile LMM model

Usage

profileGLMM_preprocess(
  regType,
  covList,
  dataframe,
  nC,
  intercept = list(FE = TRUE, RE = TRUE, Lat = TRUE)
)

Arguments

Value

A list with

• d dictionary with [XFE,XRE,XLat,UCont,UCat,ZRE] design matrices

• [[params]] list of the parameters of the data

  • n int nb of obs

  • qFE lint, number of covariates of FE

  • nRE int, number of stat units of RE

  • qRE int, number of covariates of RE

  • qLat int, number of covariates interacting with the latent clusters

  • qUCont int, number of continuous clustering covariates

  • qUCat int, number of categorical clustering covariates

  • nC int, maximal number of clusters

• prior a list with all the specification of the default prior used

• theta a list with a default set of parameters to start the chain, drawn from the prior

• regType an int. Currently 0 for linear, 1 for probit

Examples

data("exposure_data")
exp_data = exposure_data$df
theta0 = exposure_data$theta0
covList = {}
covList$FE = c('X')
covList$RE = c('t')
covList$REunit = c('indiv')

covList$Lat = c('X')

covList$Assign$Cont = c('Exp1','Exp2')
covList$Assign$Cat = NULL

covList$Y = c('Y')
dataProfile = profileGLMM_preprocess(regType = 'linear',
                                     covList = covList,
                                     dataframe = exp_data,
                                     nC = 30,
                                     intercept = list(FE = TRUE, RE = FALSE, Lat = TRUE))

Initialize the variables for the Gibbs sampler chain

Description

This function generates initial values (theta) for all parameters in the Profile GLMM Gibbs sampler by drawing from the specified prior distributions. These initial values are crucial for starting the MCMC chain in profileGLMM_Gibbs. The initialization includes parameters for fixed effects, random effects variance, latent effects, and the profile cluster parameters (centroids, covariances, and categorical probability vectors).

Usage

theta_init(prior, params)

Arguments

Value

A list (theta) containing the sampled initialization values for the Gibbs sampler. Key elements include:

sig2:

Initial residual variance.

betaFE:

Initial fixed effects coefficients.

SigRE:

Initial random effects covariance matrix.

SigLat:

Initial latent effects covariance matrix.

gammaLat:

Initial latent effects coefficients, organized by cluster.

ClusCont:

List containing initial continuous cluster parameters (mu and Sigma).

ClusCat:

List containing initial categorical cluster parameters (pvecClus).

Examples

# Load dataProfile, the result of profileGLMM_preProcess()
data("examp")
dataProfile = examp$dataProfile
theta = theta_init(dataProfile$prior,dataProfile$params)