R/madPlot.R

In muHVT: Constructing Hierarchical Voronoi Tessellations and Overlay Heatmap for Data Analysis

#' Mean Absolute Deviation Plot
#'
#' Function to create Mean Absolute Deviation Plot
#'
#' This function plots percentage anomalies vs mean absolute deviation for test data. The plot helps in
#' deciding an optimal MAD value for the use case.
#'
#'
#' @param hvt.prediction List. A list of hvt.prediction obtained from the predictHVT
#' function.
#' @param ... The ellipsis is passed to it as additional argument. (Used internally)
#' @return Mean Absolute Deviation Plot
#' \item{mad_plot}{ggplot plot. A plot with percentage anomalies on y axis and mean absolute deviation values on xaxis. }
#' @author Shubhra Prakash <shubhra.prakash@@mu-sigma.com>
#' @seealso \code{\link{predictHVT}}
#' @importFrom magrittr %>%
#' @import ggplot2
#' @examples
#' data(USArrests)
#' #Split in train and test
#'
#' train <- USArrests[1:40,]
#' test <- USArrests[41:50,]
#'
#' hvt.results <- list()
#' hvt.results <- HVT(train, n_cells = 15, depth = 1, quant.err = 0.2,
#'                    distance_metric = "L1_Norm", error_metric = "mean",
#'                    projection.scale = 10, normalize = TRUE,
#'                    quant_method="kmeans",diagnose=TRUE)
#'
#' predictions <- predictHVT(test,hvt.results, child.level=2,mad.threshold = 0.2)
#' print(predictions$scoredPredictedData)
#' madPlot(hvt.prediction=predictions)
#' @export madPlot
#' @keywords internal

madPlot = function(hvt.prediction,
                   ...) {
  # browser()
  requireNamespace("ggplot2")       #deldir function
  requireNamespace("scales")       #deldir function
  
  qe = hvt.prediction[["predictInput"]][["quant.err"]]
  df_scored = hvt.prediction[["scoredPredictedData"]]
  qe_val = df_scored$Quant.Error
  
  
  if (qe < 0.75) {
    x_breaks = seq(qe, 0.75, length.out = 10) %>% round(2)
    MAD_list = lapply(seq(qe, 0.75, by = 0.01), function(x) {
      anom_points = sum(qe_val > x)
      max_cells = length(qe_val)
      perct_anom = (anom_points / max_cells)
      
      return(c(x, anom_points, max_cells, perct_anom))
    })
  } else if (qe > 0.75) {
    x_breaks = seq(qe, (qe + 0.75), length.out = 10) %>% round(2)
    MAD_list = lapply(seq(qe, (qe + 0.75), by = 0.01), function(x) {
      anom_points = sum(qe_val > x)
      max_cells = length(qe_val)
      perct_anom = (anom_points / max_cells)
      
      return(c(x, anom_points, max_cells, perct_anom))
    })
    
    
    
  }
  
  
  comp_nclust_anom = MAD_list %>% purrr::reduce(rbind) %>% as.data.frame()
  names(comp_nclust_anom) = c(
    "Mean_Absolute_Deviation",
    "anom_points",
    "max_cells",
    "Percentage_of_Anomalous_Points"
  )
  
  ############################################
  
  mark_x = which(comp_nclust_anom$Percentage_of_Anomalous_Points < 0.01)
  mark_x = comp_nclust_anom[mark_x[1], ]
  
  
  
  mad_plot =
    ggplot()  +
    geom_point(
      data = comp_nclust_anom,
      mapping = aes(x = Mean_Absolute_Deviation,
                    y = Percentage_of_Anomalous_Points),
      shape = "circle",
      size = 1.5,
      colour = "midnightblue"
    ) +
    geom_hline(yintercept = 0.01,
               linetype = "dashed",
               colour = "brown2") +
    geom_vline(
      xintercept = mark_x$Mean_Absolute_Deviation,
      linetype = "dashed",
      colour = "brown2"
    ) +
    # geom_segment(aes(xend=mark_x$Mean_Absolute_Deviation,
    #                  yend = mark_x$Percentage_of_Anomalous_Points,
    #                  x=mark_x$Mean_Absolute_Deviation,
    #                  y=-1
    # ),linetype="dashed",colour="brown2") +
    
    annotate(
      "text",
      x = mark_x$Mean_Absolute_Deviation + (0.03 * max(
        comp_nclust_anom$Mean_Absolute_Deviation
      )),
      y = mark_x$Percentage_of_Anomalous_Points + (0.1),
      label = mark_x$Mean_Absolute_Deviation,
      fontface = "bold"
    ) +
    annotate(
      "text",
      x = qe,
      y = 0.01 + (0.1),
      label = "1%",
      fontface = "bold"
    ) +
    
    ggplot2::theme_bw() +
    ggplot2::theme(
      plot.title = ggplot2::element_text(size = 12),
      axis.text = element_text(size = 8),
      # axis.text.x = element_text(angle = 90),
      axis.title = element_text(size = 8)
    ) +
    labs(title = "Mean Absolute Deviation Plot") +
    scale_y_continuous(labels = scales::percent_format(), limits = c(0, 1)) +
    scale_x_continuous(breaks =x_breaks)
  
  
  
  # labs(title = ,
  #      subtitle = "95% Compression, QE=varying, n_cells=1033,L1Norm,Max",
  #      caption = "Mean Absolute Deviation is non dynamic i.e constant for each(earlier version was based on max QE of cells)")
  return(mad_plot)
}