plot_roc: Plot ROC curves with confidence intervals.
In mirvie/mirmodels: Models Built by the Informatics Team at Mirvie

Description Usage Arguments Details Value Examples

Plot ROC curves for several different models (or just one) with confidence intervals calculated by pROC::ci.se() and pROC::ci.sp().

plot_roc(
  rocdf,
  conf_level = 0.95,
  aes_opts = list(conf_alpha = 0.8, roc_line_size = 1, roc_line_color = "black",
    null_line_size = 1, null_line_color = "blue", null_line_alpha = 0.5),
  quick = TRUE,
  parallel = TRUE
)

`rocdf`	A data frame with columns `response` and `predictor` (see `pROC::roc()`. To plot several models, provide a `model` column with the model names; one ROC will be calculated and plot per model name.
`conf_level`	A number in `[0, 1]`. This is passed as the `conf.level` argument to `pROC::ci.se()` and `pROC::ci.sp()`. Use `conf_level = NA` to disable plotting of confidence intervals.
`aes_opts`	A named list of aesthetic options for the plot. `conf_alpha`: Transparency of the confidence interval lines. `roc_line_size`: Thickness of the ROC line. `roc_line_color`: Color of the ROC line. `null_line_size`: Thickness of the null line. `null_line_color`: Color of the null line. `null_line_alpha`: Transparency of the null line
`quick`	A flag. With `quick = FALSE`, ROC confidence intervals are calculated with a 2,000-round bootstrap. `quick = TRUE` will use 100 rounds instead. `quick = TRUE` is fine for exploration.
`parallel`	A flag. Calculate the sensitivity and specificity confidence intervals simultaneously? This is a small speedup (<2x).

Inside this function, pROC::roc() is called with direction = "<". The easiest thing is to have rocdf$response be a vector of 0s and 1s and have rocdf$predictor be a vector of probabilities where each probability is the probability of a response of 1. If rocdf$predictor is logical, then these rocdf$predictor is the probability of a response of TRUE.

A ggplot2::ggplot().

rocdf1 <- dplyr::tibble( # decent model
  predictor = seq(0.01, 0.99, length.out = 1000),
  response = purrr::rbernoulli(length(predictor), predictor)
)
rocdf2 <- dplyr::tibble( # bad model
  predictor = seq(0.01, 0.99, length.out = 88),
  response = purrr::rbernoulli(length(predictor), 0.5)
)
rocdf12 <- dplyr::bind_rows(
  dplyr::bind_cols(rocdf1, model = "decent"),
  dplyr::bind_cols(rocdf2, model = "bad")
)
plot_roc(rocdf1,
  quick = TRUE, parallel = FALSE,
  aes_opts = list(conf_alpha = 0.3)
)
plot_roc(rocdf2,
  conf_level = NA, quick = TRUE,
  aes_opts = list(roc_line_size = 2, roc_line_color = "red")
)
plot_roc(rocdf12,
  conf_level = NA, quick = TRUE,
  aes_opts = list(null_line_size = 2, null_line_alpha = 1)
)
plot_roc(rocdf12, conf_level = 0.99, quick = TRUE)