In kcf-jackson/glmSimData: A package for simulating data from a GLM model

This package simulates data following a Generalised Linear Model(GLM) model with independent / dependent, numerical / ordinal variables.

To genereate a dataset, we need to generate the covariates and the response variables. For the covariates, the key functions are

generate_independent_covariates() and
generate_dependent_covariates(),

while for the response variable, the key functions are

generate_response() and
generate_response_with_ratio().

The detailed usage is given below.

1. Independent covariates ( `generate_independent_covariates()` )

# Independent numerical
X <- generate_independent_covariates(50, 3)  #assume standard normal
X <- generate_independent_covariates(50, 3, type = 'numerical', 
      distn = 'gaussian', mean = 5, sd = 10)  
X <- generate_independent_covariates(50, 3, type = 'numerical',
      distn = 'student-t', df = 4) 
head(X)

# Independent categorical ordinal
X2 <- generate_independent_covariates(50, 3, 'categorical', 3)
X2 <- generate_independent_covariates(50, 3, 'categorical', 3, prob = c(0.25, 0.5, 0.25))
head(X2)

2. Dependent covariates ( `generate_dependent_covariates()` )

# Dependent numerical 
X <- generate_dependent_covariates(50, 5, type = 'numerical'); X;

# Dependent categorical ordinal
X <- generate_dependent_covariates(50, 5, type = 'categorical', 3); X;

3. Response variables with beta provided ( `generate_response()` )

For the response variable, we need to specify the GLM family. If we have the coefficients of the linear predictors, then we use generate_response().

# Setup
X <- generate_independent_covariates(50, 3)
beta <- rnorm(ncol(X))
family <- binomial()

# Generate the response variable
my_data <- generate_response(X, beta, family)
head(my_data)

4. Response variables with signal-noise ratio provided ( `generate_response_with_ratio()` )

If we don't have the coefficients of the linear predictors, we can specify the desired signal-to-noise ratio(SNR) instead, and we use generate_response_with_ratio().

X <- generate_independent_covariates(50, 3)
beta <- rnorm(ncol(X))
family <- binomial()

# Generate the response variable
target_ratio <- abs(rnorm(ncol(X)))
data_model_obj <- generate_response_with_ratio(X, family, target_ratio)
str(data_model_obj)