Get performance curves — get

This function generates and saves the Receiver Operating Characteristic (ROC) curve and Precision-Recall curve based on the provided metrics. It also includes the AUC values for both curves in the plot legends.

Usage

get_curves(
  data,
  spec = "Specificity",
  sens = "Sensitivity",
  reca = "Recall",
  prec = "Precision",
  color,
  auc_roc,
  auc_prc,
  LODO = FALSE,
  file.name = NULL,
  width = 6,
  height = 6,
  roc_band = NULL,
  prc_band = NULL
)

Arguments

data: A data frame containing the prediction metrics at each threshold, as returned in compute_prediction()$Metrics (sorted by decreasing predicted probability within each curve).
spec: The name of the column containing the specificity values.
sens: The name of the column containing the sensitivity values.
reca: The name of the column containing the recall values.
prec: The name of the column containing the precision values.
color: The name of the column that identifies each curve (e.g. "model" for the output of compute_prediction(), or the column with the cohort names). Each value will have a corresponding color in the plot. Several curves (several values in this column) are only supported with LODO = TRUE.
auc_roc: A list with elements estimate, lower and upper giving the AUROC and its confidence interval, as returned in compute_prediction()$AUC$AUROC. When LODO = TRUE, each element is a vector with one value per cohort, named to match the values of the color column.
auc_prc: Same structure as auc_roc, for the AUPRC (compute_prediction()$AUC$AUPRC).
LODO: Logical. If TRUE, the function assumes the data contains stacked predictions from multiple cohorts and assigns AUROC/AUPRC per cohort (default = FALSE). auc_roc and auc_prc must then hold named vectors, with the names of the cohorts.
file.name: Optional character string added to the names of the saved plots.
width: A numeric value for the width of plot
height: A numeric value for the height of plot
roc_band: Optional data frame with columns fpr, lower, upper (as in compute_prediction()$Curve_bands$ROC). If supplied and LODO = FALSE, it is drawn as a shaded pointwise confidence band around the ROC curve.
prc_band: Optional data frame with columns recall, lower, upper (as in compute_prediction()$Curve_bands$PRC), drawn around the precision-recall curve in the same way.

Value

No return value. Saves two PDF plots in the "Results/" directory: ROC_curve_<file.name>.pdf for the ROC curve and PRC_curve_<file.name>.pdf for the Precision-Recall curve (ROC_curve.pdf and PRC_curve.pdf if file.name is NULL).

Details

The ROC curve is drawn from the point (0, 0), and the precision-recall curve from recall 0 with the precision of its first point, which are the starting points used to calculate AUROC and AUPRC.

Examples

if (FALSE) { # \dontrun{
# pred: output of compute_prediction() (classification)
get_curves(data = pred$Metrics,
           color = "model",
           auc_roc = pred$AUC$AUROC,
           auc_prc = pred$AUC$AUPRC,
           roc_band = pred$Curve_bands$ROC,
           prc_band = pred$Curve_bands$PRC,
           file.name = "Example")
} # }