Metrics 0.2.0 rootmeansquarederror peroutput

Lodestar.Metrics 0.2.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

RootMeanSquaredError.PerOutput

One root 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 — the roots Score then reduces.

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 = RootMeanSquaredError.PerOutput(yTrue, yPred, outputCount: 2);
double first = perOutput[0];    // => 0.6454…
double second = perOutput[1];   // => 1

Remarks — this is exactly the square root of each entry of MeanSquaredError.PerOutput, and it is the array Score averages. Reading it is how you find out which output the headline number is being dragged down by, and unlike the squared version its entries are in each output's own units, so they can be compared against those outputs' scales.

The trap is that they still cannot be compared against each other unless the outputs share a unit. Two outputs, one in euros and one in days, give two numbers whose ratio means nothing — which is what R2.PerOutput is for.

Applies to — net10.0, netstandard2.0.

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