Home › Conformal › Conformal prediction

The gamma interval [ŷ(1 + q_low), ŷ(1 + q_up)] around a strictly positive prediction, SplitConformalRegressor(..., conformity_score="gamma").

public static (double Lower, double Upper) GammaInterval(double prediction, ReadOnlySpan<double> scores, double alpha)

Parameters — prediction is the model's point prediction, strictly positive. scores are the calibration scores from GammaScores. alpha is the miscoverage level.

Returns — the interval's two bounds, either of them infinite when the calibration set is too small for its side's level.

Exceptions — ArgumentException when scores is empty or holds a NaN; ArgumentOutOfRangeException when alpha is not strictly between 0 and 1, or prediction is not strictly positive.

Example — six calibration points at 70 % coverage.

using Lodestar.Conformal;

double[] scores = SplitConformal.GammaScores(
    [10.2, 11.5, 9.8, 12.6, 11.1, 10.4], [10.7, 10.5, 10.0, 11.8, 11.4, 11.0]);

(double lower, double upper) = SplitConformal.GammaInterval(11.0, scores, 0.3);

double low = lower;     // => 10.4
double high = upper;    // => 12.04761904761905

Remarks — it takes the scores rather than a quantile, as Interval does, because the score is not symmetric: each side reads its own quantile at α/2, as MAPIE's does, the lower one at MAPIE's own floating-point level. MAPIE's cross-conformal method="base" is this interval over the out-of-sample scores.

Applies to — net10.0, netstandard2.0.

See also — SplitConformal.GammaScores, CrossConformal.GammaInterval.