Plot Empirical ROC Curve

Plotting function to generate a receiver operator criterion (ROC) curve for binary data and binary classifiers. This function is a wrapper around geom_roc(), with window dressing for commonly used aesthetics and annotations.

Usage

plot_emp_roc(
  truth,
  predicted,
  pos_class,
  auc = TRUE,
  add = 0L,
  boot_auc = FALSE,
  adj = c(0, 0),
  auc_pos = c(0.5, 0.5),
  auc_shift = 1.25,
  auc_label = NULL,
  auc_size = 5,
  shape = NULL,
  size = 2,
  cutoff = 0.5,
  cutoff_size = 5,
  cutoff_shape = 23,
  col = 1,
  ci95 = TRUE,
  lwd = 2,
  outline = TRUE,
  boxes = TRUE,
  box_alpha = 0.35,
  debug = FALSE,
  plot_fit = FALSE,
  do_grid = TRUE
)

Arguments

truth

character(n) or factor(n). A vector of true class names. In most instances you will have to also pass a pos_class argument defining the positive/event class.

predicted

numeric(n). A numeric vector of class probabilities.

pos_class

character(1). Name of the "positive" or "event" class.

auc

logical(1). Should the AUC be printed on the plot?

add

integer(1). The position in the plotting stack indicating where to add the ROC curve relative to an existing plot. Zero indexing is used, thus add = 0L (default) refers to a new plot and add = 1L will add a ROC layer to the existing plot and correctly position the AUC annotations shifted accordingly.

boot_auc

logical(1). Should bootstrap confidence intervals of AUC be calculated?

adj

numeric(1) in [0, 1]. Coordinates that are used to align the AUC text.

auc_pos

numeric(1) in [0, 1] indicating where to place the AUC text. Must be of length == 2L, indicating the x-axis and y-axis positions, respectively. By default, the AUC will be placed slightly to the right of center of the plot.

auc_shift

The vertical (downward) shift between AUC text values if multiple ROCs are plotted.

auc_label

character(1). Adds an additional label to the AUC text, i.e. "AUC = 0.99", with label "Extra Text": "Extra Text AUC = 0.99"). Must be added individually to each plot.

auc_size

numeric(1). The size for the AUC text.

shape

numeric(1). Shape of points (between 0 and 25), similar to pch of graphics::points(). See ggplot2::geom_point().

size

numeric(1). Size of points. Similar to cex of graphics::points(). Modifying size will not affect the plot if shape is set to NULL (the default). See [geom_point())].

[geom_point())]: R:geom_point())

cutoff

numeric(1). The decision cutoff, aka operating point, for the positive classes. By default, the operating point is set to 0.5. Alternatively, choosing a negative number calculates the cutoff corresponding to the maximum perpendicular distance between the curve and the unit diagonal. The CI95 of the operating point (boxes) can be omitted by setting cutoff = NA.

cutoff_size

numeric(1) in [0, 1], the character size for the cutoff point symbol.

cutoff_shape

numeric(1). The point symbol used for the cutoff point plotted on the ROC. Defaults to a diamond (23).

col

character(1) or integer(1). Specify the colors for lines, points, bar, box, or ROC.

ci95

logical(1). Should any confidence intervals be plotted?

lwd

numeric(1). Line width (see par()).

outline

logical(1). Should black outlines be drawn around the main plot line?

boxes

logical(1). Should confidence interval boxes be drawn (showing the joint CI95 of the sensitivity and specificity confidence intervals) at the cutoff point?

box_alpha

numeric(1) in [0, 1], the shading transparency for the confidence interval box at a given cutoff.

debug

logical(1). Should debugging mode be activated? When activated, annotates the values of the plotting steps at each cutoff, prints the "positive class", prints the prediction data to the console, and various other internal objects useful for debugging.

plot_fit

logical(1). Should a maximum-likelihood (or least-squares if ML fails) fit of the curve be plotted? If plot_fit = TRUE, only the fit will be plotted, without the empirical ROC. Can also be plot_fit = "both", where a fit will be added to the empirical ROC.

do_grid

logical(1). Should grid lines be added to the ROC?

Value

A ROC curve is plotted and its corresponding AUC is returned.

Author

Michael R. Mehan, Stu Field

Examples

true <- rep(c("control", "disease"), each = 50)
pred <- withr::with_seed(8,
          c(rnorm(50, mean = 0.4, sd = 0.2),    # control predictions
            rnorm(50, mean = 0.6, sd = 0.2))    # disease predictions
        )

plot_emp_roc(true, pred, pos_class = "disease", col = "dodgerblue")

plot_emp_roc(true, pred, pos_class = "disease", ci95 = TRUE, boxes = FALSE,
             col = "red")


plot_emp_roc(true, pred, pos_class = "disease", ci95 = FALSE, shape = 21,
             col = "green")

plot_emp_roc(true, pred, pos_class = "disease", boot_auc = TRUE,
             col = "royalblue")


plot_emp_roc(true, pred, pos_class = "disease", plot_fit = "both",
             col = "purple")

plot_emp_roc(true, pred, pos_class = "disease", plot_fit = TRUE, ci95 = FALSE,
             auc = FALSE, cutoff = NA, col = 2, lwd = 4) # curve only; no cutoff


# Debugging with `debug = TRUE` displays
# curve points to the console
plot_emp_roc(true, pred, pos_class = "disease", debug = TRUE,
             col = "firebrick3")
#> ══ Debugging ══════════════════════════════════════════════════════════
#> ══ Values Top ═════════════════════════════════════════════════════════
#>      truth      pred
#> 79 disease 1.0752167
#> 84 disease 1.0089215
#> 68 disease 0.9278396
#> 92 disease 0.9137884
#> 86 disease 0.8430566
#> 51 disease 0.8159903
#> ══ Values Bottom ══════════════════════════════════════════════════════
#>      truth         pred
#> 14 control  0.141102184
#> 33 control  0.089221718
#> 77 disease  0.020540406
#> 46 control  0.008477501
#> 90 disease -0.002905448
#> 30 control -0.004121896
#> 9  control -0.202210335
#> ══ Parameters ═════════════════════════════════════════════════════════
#> † pos_class  ❯ disease
#> † boot_auc   ❯ FALSE
#> † outline    ❯ TRUE
#> † cutoff     ❯ 0.5
#> † add        ❯ 0
#> ═══════════════════════════════════════════════════════════════════════
#> # A tibble: 100 × 12
#>    cutoff    tp    fn    fp    tn sensitivity specificity   ppv   npv
#>     <dbl> <dbl> <dbl> <dbl> <dbl>       <dbl>       <dbl> <dbl> <dbl>
#>  1  1.08      1    49     0    50        0.02        1    1     0.505
#>  2  1.01      2    48     0    50        0.04        1    1     0.510
#>  3  0.928     3    47     0    50        0.06        1    1     0.515
#>  4  0.914     4    46     0    50        0.08        1    1     0.521
#>  5  0.843     5    45     0    50        0.1         1    1     0.526
#>  6  0.816     6    44     0    50        0.12        1    1     0.532
#>  7  0.808     7    43     0    50        0.14        1    1     0.538
#>  8  0.794     8    42     0    50        0.16        1    1     0.543
#>  9  0.790     8    42     1    49        0.16        0.98 0.889 0.538
#> 10  0.757     9    41     1    49        0.18        0.98 0.9   0.544
#> # ℹ 90 more rows
#> # ℹ 3 more variables: mcc <dbl>, perpD <dbl>, YoudenJ <dbl>
#> Please note: the 'size' and 'shape' arguments are locked
#>             when in debugging mode and cannot be modified.


# Multiple curves can be drawn on the same plot
true2 <- rep(c("control", "disease"), each = 50)
pred2 <- withr::with_seed(8,
          c(rnorm(50, mean = 0.5, sd = 0.3),
            rnorm(50, mean = 0.7, sd = 0.4))
        )
plot_emp_roc(true, pred, pos_class = "disease", col = "firebrick3") +
  plot_emp_roc(true2, pred2, pos_class = "disease",
               col = "dodgerblue", add = 1)

Usage

Arguments

Value

See also

Author

Examples