R/model_mbic.R

In stabiliser: Stabilising Variable Selection

#' model_mbic
#'
#' @name model_mbic
#'
#' @description Function to model mbic selection process on a given dataframe
#'
#' @param data a dataframe containing an outcome variable to be permuted (usually coming from nested bootstrap data)
#' @param outcome the outcome as a string (i.e. "y")
#' @param type model type, either "linear" or "logistic"
#' @keywords internal
#' @import bigstep
#' @import dplyr
#' @importFrom tibble rownames_to_column
#' @importFrom stats coef
#' @importFrom utils globalVariables
#' @importFrom stringr str_remove_all
#'

utils::globalVariables(c(".", "variable", "Estimate"))

model_mbic <- function(data, outcome, type) {
  data <- data %>%
    as.data.frame()
  # TODO all_of might not work with doparralel, perhaps change to !!outcome ? see tidymodels parralel explanation.
  y_temp <- data %>%
    select(all_of(outcome)) %>%
    as.matrix()

  x_temp <- data %>%
    select(-all_of(outcome))

  bigstep_prepped <- bigstep::prepare_data(y_temp, x_temp, verbose = FALSE, type = type)

  bigstep_prepped %>%
    reduce_matrix(minpv = 0.01) %>%
    fast_forward(crit = mbic) %>%
    multi_backward(crit = mbic) %>%
    summary() %>%
    stats::coef() %>%
    as.data.frame() %>%
    rownames_to_column(., var = "variable") %>%
    mutate(variable = str_remove_all(variable, "`")) %>%
    filter(
      !grepl("Xm[, -1]", variable)
    ) %>%
    rename(estimate = Estimate) %>%
    select(variable, estimate)
}