carts (version 0.1.0)

estimate_covar_model_full_cond: Full conditional covariate simulation model

Description

Estimates a full conditional model to approximate the joint distribution of covariate data. Each factor \(p(x_i | x_1, \dots x_{i-1})\) is modelled with a glm, with mean \(E[x_i | x_1, \dots x_{i-1}] = g^{-1}(\beta_0 + \sum_{j=1}^{i-1}\beta_j x_j)\). The parametric distribution of each factor is either derived from the column type (see derive_covar_distribution) or specified by cond.dist.

Usage

estimate_covar_model_full_cond(data, cond.dist = NULL)

Value

lava::lvm object with estimated coefficients

Arguments

data

Covariate data.table

cond.dist

list with random generator functions for the conditional distribution of each covariate

Author

Benedikt Sommer

Examples

data <- data.table::data.table(
y = as.factor(rbinom(1e3, size = 1, prob=0.1))
)

# infer distribution of y from column type
m.est <- estimate_covar_model_full_cond(data)
y <- sample_covar_parametric_model(1e4, m.est)$y |> as.integer() - 1
print(mean(y))

# specify distribution of y
m.est <- estimate_covar_model_full_cond(
  data, cond.dist = list(y = binomial.lvm)
)
y <- sample_covar_parametric_model(1e4, m.est)$y |> as.integer() - 1
print(mean(y))