R/PDP.R

In MattKelliher-Gibson/ModelEvalr: Tools for Evaluating and Comparing Models

#' Plots Partial Dependency Plots (PDP)
#'
#' \code{PDP} plots the Partial Dependency Plots of varaibles in a model object
#'
#' @param model Model Object (curently only gbm object tested)
#' @param iter numberic number of tree to be used from gbm object
#' @param var.name independent variable to be plotted
#' @param xlim numberic X-axis limit
#' @param pts numberic number of points used in graph. Default is \code{4000}
#' @param ylab character Y-axis title. Default is \code{"P(Conversion)"}
#' @param leg character for legend location. Default \code{"right"}
#' @param title chatacer title for plot
#' @return plot of predicted partial dependecy
#'
#' @export

PDP <- function(model, iter, var.name, xlim = NULL, pts = 4000, ylab = "P(Conversion)", leg = "right", title = NULL){

  t1 <- match(var.name, model[["var.names"]])

  t2 <- gbm::plot.gbm(model, t1, iter, return.grid = T, cont = pts)

  t2$y <- 1/(1+exp(-t2$y))

  t3 <- round(max(t2$y)-min(t2$y), 3)*100

  names(t2)[2] <- ylab

  plot(t2, type = "l", xlim = NULL, main = "", las = 2, xlab = "")

  legend(x=leg, legend = bquote(delta == .(t3)*"%"))
}

#' @export
#' @rdname PDP
PDP_gg <- function(data, model, var, iter, points = 4000){

  d1 <- match(var, model[["var.names"]])
  d2 <- gbm::plot.gbm(model, d1, iter, return.grid = TRUE, cont = points)
  d2$y2 <- 1/(1+exp(-d2$y))
  names(d2)[1] <- "Variable"

  d3 <- round(max(d2$y2) - min(d2$y2), 3) * 100

  d4 <- ggplot2::ggplot(d2, mapping = ggplot2::aes(x = Variable, y = y2))
  plot1 <- d4 + ggplot2::geom_line() + ggplot2::labs(title = paste("PDP Plot of", var), x = var, y = "P(Churn)") + ggplot2::geom_text(x = median(d2$Variable), y = max(d2$y2), label = paste0("\u0394", " = ", d3, "%"))


  data$VARIABLE <- data[[var]]
  d5 <- ggplot2::ggplot(data, mapping = ggplot2::aes(VARIABLE))
  plot2 <- d5 + ggplot2::geom_histogram() + ggplot2::labs(title = paste("Histogram of", var), y = "Total", x = var)
  gridExtra::grid.arrange(plot1, plot2, ncol = 2)
}