ai.djl.training

Class DefaultTrainingConfig

java.lang.Object

ai.djl.training.DefaultTrainingConfig

All Implemented Interfaces:

TrainingConfig

public class DefaultTrainingConfig
extends java.lang.Object
implements TrainingConfig

DefaultTrainingConfig is an implementation of the TrainingConfig interface.

Constructor Summary

Constructors

Constructor and Description
DefaultTrainingConfig(Loss loss) Creates an instance of DefaultTrainingConfig with the given Loss.

Method Summary

Modifier and Type	Method and Description
DefaultTrainingConfig	addEvaluator(Evaluator evaluator) Adds an Evaluator that needs to be computed during training.
<T extends Evaluator> DefaultTrainingConfig	addEvaluators(java.util.Collection<T> evaluators) Adds multiple Evaluators that needs to be computed during training.
DefaultTrainingConfig	addTrainingListeners(TrainingListener... listeners) Adds TrainingListeners for training.
Device[]	getDevices() Gets the Device that are available for computation.
java.util.List<Evaluator>	getEvaluators() Returns the list of Evaluators that should be computed during training.
java.util.concurrent.ExecutorService	getExecutorService() Gets the ExecutorService for parallelization.
ai.djl.util.PairList<Initializer,java.util.function.Predicate<Parameter>>	getInitializers() Gets a list of Initializer and Predicate to initialize the parameters of the model.
Loss	getLossFunction() Gets the Loss function to compute the loss against.
Optimizer	getOptimizer() Gets the Optimizer to use during training.
java.util.List<TrainingListener>	getTrainingListeners() Returns the list of TrainingListeners that should be used during training.
DefaultTrainingConfig	optDevices(Device[] devices) Sets the array of Device available for training.
DefaultTrainingConfig	optExecutorService() Sets the ExecutorService with the global ForkJoinPool.commonPool().
DefaultTrainingConfig	optExecutorService(java.util.concurrent.ExecutorService executorService) Sets the ExecutorService to train with multiple threads.
DefaultTrainingConfig	optInitializer(Initializer initializer, Parameter.Type type) Sets the Initializer to use for the parameters (default from paper).
DefaultTrainingConfig	optInitializer(Initializer initializer, java.util.function.Predicate<Parameter> predicate) Sets the Initializer to use for the parameters (default from paper).
DefaultTrainingConfig	optInitializer(Initializer initializer, java.lang.String name) Sets the Initializer to use for the parameters (default from paper).
DefaultTrainingConfig	optOptimizer(Optimizer optimizer) Sets the Optimizer used during training (default Adam).

Methods inherited from class java.lang.Object

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

Constructor Detail

DefaultTrainingConfig

public DefaultTrainingConfig(Loss loss)

Creates an instance of DefaultTrainingConfig with the given Loss. DefaultTrainingConfig creates a default TrainingConfig, Adam as optimiser, and the given Loss. The evaluators and listeners are left to the user's discretion.

Parameters:

loss - the loss to use for training

Method Detail

optInitializer

public DefaultTrainingConfig optInitializer(Initializer initializer,
                                            Parameter.Type type)

Sets the Initializer to use for the parameters (default from paper).

Parameters:

initializer - the initialer to use for the parameters

type - the Parameter.Type of the parameters

Returns:

this DefaultTrainingConfig

optInitializer

public DefaultTrainingConfig optInitializer(Initializer initializer,
                                            java.lang.String name)

Sets the Initializer to use for the parameters (default from paper).

Parameters:

initializer - the initialer to use for the parameters

name - the name of the parameter

Returns:

this DefaultTrainingConfig

optInitializer

public DefaultTrainingConfig optInitializer(Initializer initializer,
                                            java.util.function.Predicate<Parameter> predicate)

Sets the Initializer to use for the parameters (default from paper).

Parameters:

initializer - the initialer to use for the parameters

predicate - the predicate to identify parameter

Returns:

this DefaultTrainingConfig

optDevices

public DefaultTrainingConfig optDevices(Device[] devices)

Sets the array of Device available for training.

Parameters:

devices - an array of devices to be set

Returns:

this DefaultTrainingConfig

optOptimizer

public DefaultTrainingConfig optOptimizer(Optimizer optimizer)

Sets the Optimizer used during training (default Adam).

Parameters:

optimizer - the optimizer to be set

Returns:

this DefaultTrainingConfig

optExecutorService

public DefaultTrainingConfig optExecutorService()

Sets the ExecutorService with the global ForkJoinPool.commonPool().

Returns:

this DefaultTrainingConfig

optExecutorService

public DefaultTrainingConfig optExecutorService(java.util.concurrent.ExecutorService executorService)

Sets the ExecutorService to train with multiple threads.

Parameters:

executorService - the executor service

Returns:

this DefaultTrainingConfig

addEvaluators

public <T extends Evaluator> DefaultTrainingConfig addEvaluators(java.util.Collection<T> evaluators)

Adds multiple Evaluators that needs to be computed during training.

Type Parameters:

T - the type of evaluator to be added

Parameters:

evaluators - the evaluators to be added

Returns:

this DefaultTrainingConfig

addEvaluator

public DefaultTrainingConfig addEvaluator(Evaluator evaluator)

Adds an Evaluator that needs to be computed during training.

Parameters:

evaluator - the evaluator to be added

Returns:

this DefaultTrainingConfig

addTrainingListeners

public DefaultTrainingConfig addTrainingListeners(TrainingListener... listeners)

Adds TrainingListeners for training.

Parameters:

listeners - the TrainingListeners to add

Returns:

this DefaultTrainingConfig

getDevices

public Device[] getDevices()

Gets the Device that are available for computation.

This is necessary for a Trainer as it needs to know what kind of device it is running on, and how many devices it is running on.

Specified by:

getDevices in interface TrainingConfig

Returns:

an array of Device

getInitializers

public ai.djl.util.PairList<Initializer,java.util.function.Predicate<Parameter>> getInitializers()

Gets a list of Initializer and Predicate to initialize the parameters of the model.

Specified by:

getInitializers in interface TrainingConfig

Returns:

an Initializer

getOptimizer

public Optimizer getOptimizer()

Gets the Optimizer to use during training.

Specified by:

getOptimizer in interface TrainingConfig

Returns:

an Optimizer

getLossFunction

public Loss getLossFunction()

Gets the Loss function to compute the loss against.

Specified by:

getLossFunction in interface TrainingConfig

Returns:

a Loss function

getExecutorService

public java.util.concurrent.ExecutorService getExecutorService()

Description copied from interface: TrainingConfig

Gets the ExecutorService for parallelization.

Specified by:

getExecutorService in interface TrainingConfig

Returns:

an ExecutorService

getEvaluators

public java.util.List<Evaluator> getEvaluators()

Returns the list of Evaluators that should be computed during training.

Specified by:

getEvaluators in interface TrainingConfig

Returns:

a list of Evaluators

getTrainingListeners

public java.util.List<TrainingListener> getTrainingListeners()

Returns the list of TrainingListeners that should be used during training.

Specified by:

getTrainingListeners in interface TrainingConfig

Returns:

a list of TrainingListeners