Evaluation

classin packageklyn.math.nn

public class Evaluation

Evaluation

Sample-weighted supervised metrics, independent of batch partitioning. Classification accuracy is a fraction in [0, 1], not a percentage. Regression reports loss only and never fabricates an accuracy value.

Example

print(model.evaluate(xTest, yTest).accuracy)

Properties

Modifier and Type	Member	Description
public readonly property	accuracy `accuracy as Double`	Fraction of correctly classified samples.
public readonly property	hasAccuracy `hasAccuracy as Boolean`	Whether accuracy was evaluated for classification.
public readonly property	loss `loss as Double`	Mean per-sample loss.
public readonly property	samples `samples as ULong`	Number of samples contributing to the metrics.

Constructors

Modifier and Type	Member	Description
public	Evaluation `Evaluation(totalLoss as Double, correct as ULong, samples as ULong)`	Constructs metrics from accumulated loss and correct predictions.
public	Evaluation `Evaluation(totalLoss as Double, samples as ULong)`	Constructs loss-only regression metrics.

Methods

Modifier and Type	Member	Description
public static	fromMean `fromMean(meanLoss as Double, samples as ULong, correct as ULong = 0ul, hasAccuracy as Boolean = false) as Evaluation`	Builds metrics from an already reduced mean without multiplying it by the sample count, which could overflow even when the mean is finite.