SeaVal (version 1.1.1)

ROC_curve: ROC curve for tercile forecasts

Description

Creates ROC curves from a data table containing tercile forecasts. It wraps roc_curve_vec. By default, 4 ROC-curves are drawn, one for each the prediction of above-, normal- and below-values, plus one for all forecasts together. You can provide a 'by' argument to obtain separate ROC-curves for different values of the by-columns. E.g., when your data table contains a column named 'season', you can set by = 'season'. Then, the function will output a list of 16 ROC-curvess, 4 for each season.

Usage

ROC_curve(
  dt,
  f = c("below", "normal", "above"),
  o = tc_cols(dt),
  by = NULL,
  pool = setdiff(dimvars(dt), by),
  interpolate = TRUE,
  dim.check = TRUE
)

Value

A list of gg objects which can be plotted by ggpubr::ggarrange (for example)

Arguments

dt

Data table containing tercile forecasts

f

column names of the prediction columns

o

column name of the observation column

by

column names of grouping variables. Default is to not group.

pool

column names of pooling variables (used for the dimension check). Default is all dimvars.

interpolate

Logical. If TRUE, the curve connects the dots making up the ROC curve (which looks nicer), if not a step function is drawn (which is closer to the mathematical definition of the ROC curve).

dim.check

Logical. If TRUE, the function checks whether the columns in by and pool span the entire data table.

Examples

dt = data.table(below = c(0.5,0.3,0),
                normal = c(0.3,0.3,0.7),
                above = c(0.2,0.4,0.3),
                tc_cat = c(-1,0,0),
                lon = 1:3)
print(dt)
pp = ROC_curve(dt)
if(interactive()) plot(pp)