org.encog.neural.networks.training.concurrent

Class ConcurrentTrainingManager

java.lang.Object

org.encog.neural.networks.training.concurrent.ConcurrentTrainingManager

All Implemented Interfaces:

java.lang.Runnable

public final class ConcurrentTrainingManager
extends java.lang.Object
implements java.lang.Runnable

Concurrent training manager. This class allows you to queue up network training tasks to be executed either by the CPU cores or OpenCL devices. This allows the CPU/GPU to train neural networks at the same time.

Method Summary

Modifier and Type	Method and Description
void	addPerformer(ConcurrentTrainingPerformer performer) Add a performer.
void	addTrainingJob(TrainingJob job) Add a training job.
void	clearPerformers() Clear all of the performers.
void	clearQueue() Clear the workload.
void	detectPerformers() Detect performers.
void	detectPerformers(boolean splitCores) Detect performers.
static ConcurrentTrainingManager	getInstance()
boolean	isSingleThreaded()
void	jobDone(long time, ConcurrentTrainingPerformerCPU perf) Wait for job to finish.
void	join() Wait for all tasks to finish.
void	run() Perform the training.
void	setReport(StatusReportable report) Setup the object to report status to.
void	setSingleThreaded(boolean singleThreaded)
void	start() Start the manager.
java.lang.String	toString()
ConcurrentTrainingPerformer	waitForFreePerformer() Wait for a free performer.

Methods inherited from class java.lang.Object

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

Method Detail

getInstance

public static ConcurrentTrainingManager getInstance()

Returns:

The singleton instance.

addPerformer

public void addPerformer(ConcurrentTrainingPerformer performer)

Add a performer.

Parameters:

performer - The performer to add.

addTrainingJob

public void addTrainingJob(TrainingJob job)

Add a training job.

Parameters:

job - The training job to add.

clearPerformers

public void clearPerformers()

Clear all of the performers.

clearQueue

public void clearQueue()

Clear the workload.

detectPerformers

public void detectPerformers()

Detect performers. Create one performer for each OpenCL device, and another for the CPU's. If there is an OpenCL device already for the CPU, do not create another CPU performer.

detectPerformers

public void detectPerformers(boolean splitCores)

Detect performers. Create one performer for each OpenCL device, and another for the CPU's. If there is an OpenCL device already for the CPU, do not create another CPU performer.

Parameters:

splitCores - True, if a CPU performer should be created for each core.

isSingleThreaded

public boolean isSingleThreaded()

Returns:

True, if single threaded.

jobDone

public void jobDone(long time,
                    ConcurrentTrainingPerformerCPU perf)

Wait for job to finish.

Parameters:

time - The time to wait.

perf - The performer.

join

public void join()

Wait for all tasks to finish.

run

public void run()

Perform the training.

Specified by:

run in interface java.lang.Runnable

setReport

public void setReport(StatusReportable report)

Setup the object to report status to.

Parameters:

report - The object to report status to.

setSingleThreaded

public void setSingleThreaded(boolean singleThreaded)

start

public void start()

Start the manager.

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object

waitForFreePerformer

public ConcurrentTrainingPerformer waitForFreePerformer()

Wait for a free performer.

Returns:

The free performer.