Add (or override) fill aesthetic to a ggplot2 autoplot function

Question

I would like to add a fill aesthetic to an autoplot function. Specifically, to the autoplot.conf_mat function from the yardstick package.

library(tidyverse)
library(tidymodels)

data("hpc_cv")

cm <- hpc_cv %>%
  filter(Resample == "Fold01") %>%
  conf_mat(obs, pred)
cm
#>           Truth
#> Prediction  VF   F   M   L
#>         VF 166  33   8   1
#>         F   11  71  24   7
#>         M    0   3   5   3
#>         L    0   1   4  10

autoplot(cm, type = "mosaic") 

^{Created on 2020-11-20 by the reprex package (v0.3.0)}

QUESTION: How can I add a fill aesthetic to either the Truth or Prediction factor? That is, add some color to the different categories.

I have tried + scale_fill_manual and other variants, but I do not believe a fill aesthetic is being utilized in the autoplot call.

Answer 1

I'll show you two ways to get the result you want:

• with a custom function
• with autoplot

---

1. WITH A CUSTOM FUNCTION

The function that autoplot calls is cs_mosaic .

You can rewrite that function to add the labels you need, in this way:

cm_mosaic_fill <- function(x, fill){
 
 `%+%` <- ggplot2::`%+%`
 cm_zero <- (as.numeric(x$table == 0)/2) + x$table
 x_data <- yardstick:::space_fun(colSums(cm_zero), 200)
 full_data_list <- purrr::map(seq_len(ncol(cm_zero)), 
                              ~yardstick:::space_y_fun(cm_zero, .x, x_data))
 full_data <- dplyr::bind_rows(full_data_list)
 y1_data <- full_data_list[[1]]
 tick_labels <- colnames(cm_zero)
 
 ####### { EDIT: add fill
 full_data$Predicted <- tick_labels
 full_data$Truth     <- rep(tick_labels, each = nrow(x_data))
 ####### }

 ggplot2::ggplot(full_data) %+% 
  ggplot2::geom_rect(ggplot2::aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax, 

  ####### { EDIT: add fill
                     fill = !!enquo(fill))) %+%
  ####### }

  ggplot2::scale_x_continuous(breaks = (x_data$xmin + x_data$xmax)/2, labels = tick_labels) %+% 
  ggplot2::scale_y_continuous(breaks = (y1_data$ymin + y1_data$ymax)/2, labels = tick_labels) %+% 
  ggplot2::labs(y = "Predicted", x = "Truth") %+% 
  ggplot2::theme(panel.background = ggplot2::element_blank())
 
}

cm_mosaic_fill(cm, Truth)

cm_mosaic_fill(cm, Predicted)

---

2. WITH AUTOPLOT

You can do it directly with autoplot too, but it's tricky and not really readable or maintainable in my opinion.

autoplot(cm, type = "mosaic") + aes(fill = rep(colnames(cm$table), each = ncol(cm$table))) + labs(fill = "Truth")

autoplot(cm, type = "mosaic") + aes(fill = rep(colnames(cm$table), ncol(cm$table))) + labs(fill = "Predicted")