Home › Conformal › Conformal prediction

The gamma calibration scores of a regressor: (y − ŷ) / ŷ, for a strictly positive target.

public static double[] GammaScores(ReadOnlySpan<double> yTrue, ReadOnlySpan<double> yPredicted)

Parameters — yTrue are the observed values and yPredicted the model's predictions for them, all strictly positive.

Returns — one signed score per point, in the input's order. Hand them to GammaInterval, or to CrossConformal.GammaInterval when the predictions are out of sample.

Exceptions — ArgumentException when the spans have different lengths; ArgumentOutOfRangeException when a value or a prediction is zero, negative or NaN.

Example — a relative error, signed.

using Lodestar.Conformal;

double[] scores = SplitConformal.GammaScores([10.2, 12.6], [10.7, 11.8]);

double under = scores[0];    // => -0.04672897196261683
double over = scores[1];     // => 0.06779661016949143

Remarks — MAPIE's GammaConformityScore: where the error grows with the target — prices, durations, counts — a relative score gives an interval proportional to the prediction, and a signed one lets the two sides differ. A non-positive value is refused, as MAPIE refuses it, since the score's support is the positive reals.

Applies to — net10.0, netstandard2.0.

See also — SplitConformal.GammaInterval, SplitConformal.AbsoluteResiduals.