Metrics d2pinball peroutput

Development build. This page describes main, not a released package. The latest published Lodestar.Metrics is 0.3.0 — read its documentation.

D2Pinball.PerOutput

The explained fraction, one number per output column — d2_pinball_score(…, multioutput="raw_values").

public static double[] PerOutput(ReadOnlySpan<double> yTrue, ReadOnlySpan<double> yPred, double alpha = 0.5, int outputCount = 1, ReadOnlySpan<double> sampleWeight = default, ZeroDivision zeroDivision = ZeroDivision.NaN)

Parameters — yTrue and yPred are the true and predicted values, row-major. alpha is the quantile being scored, in [0, 1]. outputCount is how many outputs each row holds. sampleWeight is one weight per row. zeroDivision decides the answer for fewer than two samples; see D2Pinball.Score.

Returns — a fresh double[] of outputCount entries, in column order. Each is what D2Pinball.Score would return for that column on its own, quantile and all — the baseline is computed per column, not once for the whole matrix.

Exceptions — ArgumentException when a length disagrees with the shape, the input is empty, or it holds a non-finite value. ArgumentOutOfRangeException when outputCount is below one, or alpha is outside [0, 1]. UndefinedMetricException when there are fewer than two samples and zeroDivision is ZeroDivision.Throw.

Example — three samples over two outputs, scored at the median.

using Lodestar.Metrics;

double[] truth = [0.5, 1.0, 1.0, 1.0, 7.0, -6.0];
double[] predicted = [0.0, 2.0, -1.0, 2.0, 8.0, -5.0];

double[] columns = D2Pinball.PerOutput(truth, predicted, alpha: 0.5, outputCount: 2);
double second = columns[1];  // => 0.5714…

Averaging the two gives what D2Pinball.Score returns with no outputWeights — 0.5164… here.

Applies to — net10.0, netstandard2.0.

See also — D2Pinball.Score, D2AbsoluteError.PerOutput, the Python equivalence table.