Metrics meansquarederror peroutput

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

MeanSquaredError.PerOutput

One mean squared error per output, unreduced.

public static double[] PerOutput(ReadOnlySpan<double> yTrue, ReadOnlySpan<double> yPred, int outputCount = 1, ReadOnlySpan<double> sampleWeight = default)

Parameters — yTrue and yPred are the true and predicted values, row-major. outputCount is how many outputs each row holds, and sampleWeight weights the rows.

Returns — a fresh double[] of outputCount entries, in column order.

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.

Example — three samples, two outputs.

using Lodestar.Metrics;

double[] yTrue = [0.5, 1.0, -1.0, 1.0, 7.0, -6.0];
double[] yPred = [0.0, 2.0, -1.0, 2.0, 8.0, -5.0];

double[] perOutput = MeanSquaredError.PerOutput(yTrue, yPred, outputCount: 2);
double first = perOutput[0];    // => 0.4166…
double second = perOutput[1];   // => 1

Remarks — the per-output array is where a multioutput model is actually diagnosed, because squared errors in different units cannot be averaged into anything meaningful. Two outputs, one in euros and one in days, give a Score in "euros-squared and days-squared", which is not a quantity.

The trap follows from that: outputWeights on Score is often used to fix it, and it does not. Weighting a euro-squared against a day-squared still leaves a number with no units. If the outputs are on different scales, the fix is to normalise the targets or to report R2.PerOutput, which is unitless by construction.

Applies to — net10.0, netstandard2.0.

See also — MeanSquaredError.Score, RootMeanSquaredError.PerOutput, R2.PerOutput, the Python equivalence table.