RROC Curves Software

This is R code for calculating and drawing RROC curves (ROC curves for regression models) and their convex hulls.

For information about RROC curves (and ROC analysis for regression), see the following paper:

The program calculates and draws:

RROC curves
- Convex hull of several curves.
- Their areas
- Isometrics
- Dominance intervals for several curves
RCOST curves
- REC curves
- Their areas
It also includes code to deal with a complete work-to-deployment process, using a real dataset.

Available versions:

Implemented functions:

plot_RROCspace(...): Plots the underlying RROC space (initially empty)
RROC_curve(...) : Returns a RROC curve of a regression model and (optionally) plots it on a previously created RROC space
RROC_convexhull(...): Returns convex hull of a set of points in RROC space and (optionally) plots it on a previously created RROC space
calibrate_regressor(...): Calibrates a regression model by adding a constant shift to get 0 bias (overpredictions = underpredictions).
RROC_intercept(...): Intercepts a cost context (alpha) on a RROC curve (assumes the curve is convex)
RROC_intercept_many(...): Intercepts a cost context (alpha) on several RROC curve (as RROC_intercept_many) but for several curves in a list
RROC_dominance_intervals(...): Finds the models that dominate (and their ranges) for several curves in a list
REC_space(...): Plots the underlying REC space (use this just before plotting any model)
REC_curve(...) : Returns a REC curve of a regression model and (optionally) plots it on a previously created REC space
loss_linlin(...): Returns the asymmetric absolute error loss (linlin)
optimal_shift_linlin (...): Returns the best shift for a regression model (wrt. the linlin loss determined by an alpha)