TrainingHistory

classin packageklyn.math.nn

public class TrainingHistory

TrainingHistory

Training and optional validation metrics after each completed epoch of one fit call. Training metrics describe predictions made as parameters evolved within the epoch; validation metrics use the fixed parameters at the end of that epoch. Plot-ready metric series are extracted together on first access and cached until another epoch is recorded. Repeated reads return the same immutable snapshots without traversing the history again.

Example

print(history.epochs[0].loss)

Properties

Modifier and Type	Member	Description
public readonly property	epochNumbers `epochNumbers as IList<Double>`	Immutable one-based epoch numbers as Double values, suitable for chart axes.
public readonly property	epochs `epochs as IList<Evaluation>`	Immutable snapshot of completed epoch metrics, cached until recording.
public readonly property	trainingAccuracy `trainingAccuracy as IList<Double>`	Immutable training accuracy for each completed epoch, as fractions [0, 1].
public readonly property	trainingLoss `trainingLoss as IList<Double>`	Immutable sample-weighted training loss for each completed epoch.
public readonly property	validationAccuracy `validationAccuracy as IList<Double>`	Immutable validation accuracy, as fractions [0, 1].
public readonly property	validationEpochs `validationEpochs as IList<Evaluation>`	Immutable snapshot of validation metrics, evaluated with fixed parameters after each epoch and dropout disabled.
public readonly property	validationLoss `validationLoss as IList<Double>`	Immutable sample-weighted validation loss, aligned with epochNumbers when validation is enabled.