R/mod_penalized.R

In cottinlola/modeleVariablesAP: Apply models to all variables

#' Fits a GLM penalized model using k-fold cross-validation to tune the
#' lambda hyperparameter
#'
#' @param data A numeric dataset in matrix-convertible format
#' @param x_names A character vector of explainatory variables names
#' @param y_name A character naming the variable to explain
#' @param alpha A numeric to compromise between ridge (alpha=0) and lasso
#'              (alpha=1, default) penalization
#' @param type_measure  A character naming the error metric used for the
#'                      cross-validation (defaults to "mse")
#'
#' @return The fitted model
#'
#' @export
#'
#' @example
#' mod <- mod_penalized(data, y_name = "SUBEx")
#'
mod_penalized <- function(data, x_names = character(), y_name,
                          alpha = 1, type_measure = "mse") {
  x <- data
  y <- data[, y_name]
  if (length(x_names) != 0) {
    x <- x[, x_names]
  }
  x <- data_numeric_only(x)
  if (y_name %in% colnames(x)) {
    x <- x[, -which(colnames(x) == y_name)]
  }
  mod_en <- glmnet::cv.glmnet(x = as.matrix(x), y = y, alpha = alpha,
                              type.measure = type_measure)
  mod_en$glmnet.fit$cvm <- mod_en$cvm
  mod_en$glmnet.fit$nzero <- mod_en$nzero
  mod_en$glmnet.fit$lambda.min <- mod_en$lambda.min # nolint
  mod_en$glmnet.fit$lambda.1se <- mod_en$lambda.1se # nolint
  return(mod_en$glmnet.fit)
}