ai.djl.training.listener

Class EvaluatorTrainingListener

java.lang.Object

ai.djl.training.listener.TrainingListenerAdapter

ai.djl.training.listener.EvaluatorTrainingListener

All Implemented Interfaces:

TrainingListener

public class EvaluatorTrainingListener
extends TrainingListenerAdapter

TrainingListener that records evaluator results.

Results are recorded for the following stages:

• TRAIN_EPOCH - This accumulates for the whole epoch and is recorded to a metric at the end of the epoch
• TRAIN_PROGRESS - This accumulates for progressUpdateFrequency batches and is recorded to a metric at the end
• TRAIN_ALL - This does not accumulates and records every training batch to a metric
• VALIDATE_EPOCH - This accumulates for the whole validation epoch and is recorded to a metric at the end of the epoch

The training and validation evaluators are saved as metrics with names that can be found using metricName(Evaluator, String). The validation evaluators are also saved as model properties with the evaluator name.

Nested Class Summary

Nested classes/interfaces inherited from interface ai.djl.training.listener.TrainingListener

TrainingListener.BatchData, TrainingListener.Defaults

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.String	TRAIN_ALL
static java.lang.String	TRAIN_EPOCH
static java.lang.String	TRAIN_PROGRESS
static java.lang.String	VALIDATE_EPOCH

Constructor Summary

Constructors

Constructor and Description
EvaluatorTrainingListener() Constructs an EvaluatorTrainingListener that updates the training progress the default frequency.
EvaluatorTrainingListener(int progressUpdateFrequency) Constructs an EvaluatorTrainingListener that updates the training progress the given frequency.

Method Summary

Modifier and Type	Method and Description
java.util.Map<java.lang.String,java.lang.Float>	getLatestEvaluations() Returns the latest evaluations.
static java.lang.String	metricName(Evaluator evaluator, java.lang.String stage) Returns the metric created with the evaluator for the given stage.
void	onEpoch(Trainer trainer) Listens to the end of an epoch during training.
void	onTrainingBatch(Trainer trainer, TrainingListener.BatchData batchData) Listens to the end of training one batch of data during training.
void	onTrainingBegin(Trainer trainer) Listens to the beginning of training.
void	onValidationBatch(Trainer trainer, TrainingListener.BatchData batchData) Listens to the end of validating one batch of data during validation.

Methods inherited from class ai.djl.training.listener.TrainingListenerAdapter

onTrainingEnd

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

TRAIN_EPOCH

public static final java.lang.String TRAIN_EPOCH

See Also:

Constant Field Values

TRAIN_PROGRESS

public static final java.lang.String TRAIN_PROGRESS

See Also:

Constant Field Values

TRAIN_ALL

public static final java.lang.String TRAIN_ALL

See Also:

Constant Field Values

VALIDATE_EPOCH

public static final java.lang.String VALIDATE_EPOCH

See Also:

Constant Field Values

Constructor Detail

EvaluatorTrainingListener

public EvaluatorTrainingListener()

Constructs an EvaluatorTrainingListener that updates the training progress the default frequency.

Current default frequency is every 5 batches.

EvaluatorTrainingListener

public EvaluatorTrainingListener(int progressUpdateFrequency)

Constructs an EvaluatorTrainingListener that updates the training progress the given frequency.

Parameters:

progressUpdateFrequency - the number of batches to accumulate an evaluator before it is stable enough to output

Method Detail

onEpoch

public void onEpoch(Trainer trainer)

Listens to the end of an epoch during training.

Specified by:

onEpoch in interface TrainingListener

Overrides:

onEpoch in class TrainingListenerAdapter

Parameters:

trainer - the trainer the listener is attached to

onTrainingBatch

public void onTrainingBatch(Trainer trainer,
                            TrainingListener.BatchData batchData)

Listens to the end of training one batch of data during training.

Specified by:

onTrainingBatch in interface TrainingListener

Overrides:

onTrainingBatch in class TrainingListenerAdapter

Parameters:

trainer - the trainer the listener is attached to

batchData - the data from the batch

onValidationBatch

public void onValidationBatch(Trainer trainer,
                              TrainingListener.BatchData batchData)

Listens to the end of validating one batch of data during validation.

Specified by:

onValidationBatch in interface TrainingListener

Overrides:

onValidationBatch in class TrainingListenerAdapter

Parameters:

trainer - the trainer the listener is attached to

batchData - the data from the batch

onTrainingBegin

public void onTrainingBegin(Trainer trainer)

Listens to the beginning of training.

Specified by:

onTrainingBegin in interface TrainingListener

Overrides:

onTrainingBegin in class TrainingListenerAdapter

Parameters:

trainer - the trainer the listener is attached to

metricName

public static java.lang.String metricName(Evaluator evaluator,
                                          java.lang.String stage)

Returns the metric created with the evaluator for the given stage.

Parameters:

evaluator - the evaluator to read the metric from

stage - one of TRAIN_EPOCH, TRAIN_PROGRESS, or VALIDATE_EPOCH

Returns:

the metric name to use

getLatestEvaluations

public java.util.Map<java.lang.String,java.lang.Float> getLatestEvaluations()

Returns the latest evaluations.

The latest evaluations are updated on each epoch.

Returns:

the latest evaluations