R/tidy_agtboost.R

In autostats: Auto Stats

#' #' tidy agtboost
#' #'
#' #' Boosted tree regression using the  \href{https://github.com/Blunde1/agtboost}{agtboost} package.
#' #' Variable importance plot is printed along with returning the model. Noise features are eliminated from the plot.
#' #'
#' #' \href{https://arxiv.org/abs/2008.12625}{agtboost: Adaptive and Automatic Gradient Tree Boosting Computations}
#' #'
#' #'
#' #'
#' #' @param .data dataframe
#' #' @param formula formula
#' #' @param ... additional parameters to pass to \code{\link[agtboost]{gbt.train}}
#' #'
#' #' @return agtboost model of class \code{Rcpp_ENSEMBLE}
#' #' @export
#' #'
#' #' @examples
#' #'
#' #' iris %>%
#' #'tidy_formula(target = Petal.Length) -> f1
#' #'
#' #'iris %>%
#' #'  tidy_agtboost(f1)
#' tidy_agtboost <- function(.data, formula, ...){
#'
#'   formula %>%
#'     rlang::f_lhs() -> target
#'
#'   formula %>%
#'     f_formula_to_charvec(.data = .data) -> predictor_vars
#'
#'   .data %>%
#'     framecleaner::filter_missing(tidyselect::any_of(predictor_vars),
#'                                  tidyselect::any_of(target)) -> .data
#'
#'
#'
#'   .data %>%
#'     dplyr::select(tidyselect::any_of(predictor_vars)) %>%
#'     framecleaner::create_dummies(remove_most_frequent_dummy = TRUE) -> train_data
#'
#'   train_data %>%
#'   as.matrix -> train_matrix
#'
#'   .data %>%
#'     dplyr::pull(!!target) -> tg
#'
#'   tg %>%
#'     is.numeric() -> numer_tg
#'
#'   if(!numer_tg){rlang::abort("Target must be numeric")}
#'
#'   agtboost::gbt.train(y = tg, x = train_matrix, ...) -> gbt_model
#'
#'   agtboost::gbt.importance(feature_names = names(train_data), object = gbt_model)
#'
#'   gbt_model
#'
#'
#' }