Metrics meansquaredlogerror peroutput

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

MeanSquaredLogError.PerOutput

One mean squared log 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, every value above −1. 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, it holds a non-finite value, or either array holds a value at or below −1; ArgumentOutOfRangeException when outputCount is below one.

Example — the same four counts read as two samples of two outputs.

using Lodestar.Metrics;

double[] yTrue = [3.0, 5.0, 2.5, 7.0];
double[] yPred = [2.5, 5.0, 4.0, 8.0];

double[] perOutput = MeanSquaredLogError.PerOutput(yTrue, yPred, outputCount: 2);
double first = perOutput[0];    // => 0.0725…
double second = perOutput[1];   // => 0.0069…

Remarks — log units are at least the same units for every output, so unlike MeanSquaredError.PerOutput these entries can honestly be compared and averaged even when the targets are counts of different things.

The trap is the −1 rule applying to the whole span, not per output. One negative value anywhere in yTrue or yPred refuses the call, so a multioutput target with one column that can legitimately go negative cannot use this at all — split the columns and score them separately.

Applies to — net10.0, netstandard2.0.

See also — MeanSquaredLogError.Score, RootMeanSquaredLogError.PerOutput, the Python equivalence table.