org.nd4j.linalg.solvers.api

Interface TrainingEvaluator

All Superinterfaces:

Serializable

public interface TrainingEvaluator
extends Serializable

Training evaluator, used for determining early stop

Author:

Adam Gibson

Method Summary

Modifier and Type	Method and Description
double	bestLoss() The best validation loss so far
double	improvementThreshold()
int	miniBatchSize()
double	patience()
double	patienceIncrease() Amount patience should be increased when a new best threshold is hit
boolean	shouldStop(int epoch) Whether to terminate or not
int	validationEpochs() The number of epochs to test on

Method Detail

shouldStop

boolean shouldStop(int epoch)

Whether to terminate or not

Parameters:

epoch - the current epoch

Returns:

whether to terminate or not on the given epoch

improvementThreshold

double improvementThreshold()

patience

double patience()

patienceIncrease

double patienceIncrease()

Amount patience should be increased when a new best threshold is hit

Returns:

bestLoss

double bestLoss()

The best validation loss so far

Returns:

the best validation loss so far

validationEpochs

int validationEpochs()

The number of epochs to test on

Returns:

the number of epochs to test on

miniBatchSize

int miniBatchSize()