Model matrix with all levels option — modmat • activesampling

Construct a model matrix while optionally retaining all levels of factor variables (i.e., do not drop a reference level). This can be useful for LASSO-style models or other workflows where a fixed reference level is undesirable. Optionally returns a sparse model matrix.

Usage

modmat(formula, df, sparse = FALSE, all_levels = TRUE)

Arguments

formula: A model formula passed to model.matrix.
df: A data.frame providing the variables in formula.
sparse: Logical; if TRUE, return a sparse model matrix via Matrix::sparse.model.matrix.
all_levels: Logical; if TRUE, keep all levels of all factor variables (non-full-rank design). If FALSE, use default treatment contrasts which drop one level (full rank).

Value

A dense (matrix) or sparse (dgCMatrix) model matrix.

Details

This helper mirrors the behavior of a legacy script that prefixed factor and character values to produce stable column names and then supplied explicit contrast matrices to model.matrix/sparse.model.matrix.

Examples

df <- data.frame(y = rnorm(5), f = factor(c("a","b","a","c","b")), x = 1:5)
mm1 <- modmat(y ~ f + x, df, sparse = FALSE, all_levels = TRUE)
mm2 <- modmat(y ~ f + x, df, sparse = TRUE, all_levels = FALSE)