org.deeplearning4j.parallelism.factory

Class SymmetricTrainerContext

java.lang.Object

org.deeplearning4j.parallelism.factory.SymmetricTrainerContext

All Implemented Interfaces:

TrainerContext

public class SymmetricTrainerContext
extends Object
implements TrainerContext

Creates DefaultTrainer instances for use with ParallelWrapper

Author:

raver119@gmail.com

Constructor Summary

Constructors

Constructor and Description
SymmetricTrainerContext()

Method Summary

Modifier and Type	Method and Description
Trainer	create(int threadId, org.deeplearning4j.nn.api.Model model, int rootDevice, boolean useMDS, ParallelWrapper wrapper, org.deeplearning4j.nn.conf.WorkspaceMode mode, int averagingFrequency) Create a Trainer based on the given parameters
void	finalizeRound(org.deeplearning4j.nn.api.Model originalModel, org.deeplearning4j.nn.api.Model... models) This method is called at averagingFrequency
void	finalizeTraining(org.deeplearning4j.nn.api.Model originalModel, org.deeplearning4j.nn.api.Model... models) This method is called
void	init(org.deeplearning4j.nn.api.Model model, Object... args) Initialize the context

Methods inherited from class java.lang.Object

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

Constructor Detail

SymmetricTrainerContext

public SymmetricTrainerContext()

Method Detail

init

public void init(org.deeplearning4j.nn.api.Model model,
                 Object... args)

Initialize the context

Specified by:

init in interface TrainerContext

Parameters:

model -

args - the arguments to initialize with (maybe null)

create

public Trainer create(int threadId,
                      org.deeplearning4j.nn.api.Model model,
                      int rootDevice,
                      boolean useMDS,
                      ParallelWrapper wrapper,
                      org.deeplearning4j.nn.conf.WorkspaceMode mode,
                      int averagingFrequency)

Create a Trainer based on the given parameters

Specified by:

create in interface TrainerContext

Parameters:

threadId - the thread id to use for this worker

model - the model to start the trainer with

rootDevice - the root device id

useMDS - whether to use the MagicQueue or not

wrapper - the wrapper instance to use with this trainer (this refernece is needed for coordination with the ParallelWrapper 's IterationListener

Returns:

the created training instance

finalizeRound

public void finalizeRound(org.deeplearning4j.nn.api.Model originalModel,
                          org.deeplearning4j.nn.api.Model... models)

Description copied from interface: TrainerContext

This method is called at averagingFrequency

Specified by:

finalizeRound in interface TrainerContext

finalizeTraining

public void finalizeTraining(org.deeplearning4j.nn.api.Model originalModel,
                             org.deeplearning4j.nn.api.Model... models)

Description copied from interface: TrainerContext

This method is called

Specified by:

finalizeTraining in interface TrainerContext