Package cc.mallet.fst

Class ThreadedOptimizable

java.lang.Object

cc.mallet.fst.ThreadedOptimizable

All Implemented Interfaces:

Optimizable, Optimizable.ByGradientValue

public class ThreadedOptimizable
extends java.lang.Object
implements Optimizable.ByGradientValue

An adaptor for optimizables based on batch values/gradients.

Computes values, gradients for each batch in multiple threads and combines them in the end.

Author:

Gaurav Chandalia

See Also:

CRFOptimizableByBatchLabelLikelihood

Nested Class Summary

Nested classes/interfaces inherited from interface cc.mallet.optimize.Optimizable

Optimizable.ByBatchGradient, Optimizable.ByCombiningBatchGradient, Optimizable.ByGISUpdate, Optimizable.ByGradient, Optimizable.ByGradientValue, Optimizable.ByHessian, Optimizable.ByValue, Optimizable.ByVotedPerceptron

Field Summary

Fields

Modifier and Type	Field	Description
protected java.util.List<double[]>	batchCachedGradient	Gradient obtained from the optimizable for each batch
protected double[]	batchCachedValue	Value obtained from the optimizable for each batch
protected CacheStaleIndicator	cacheIndicator
protected Optimizable.ByCombiningBatchGradient	optimizable	optimizable to be parallelized
static int	SLEEP_TIME
protected InstanceList	trainingSet	Data

Constructor Summary

Constructors

Constructor	Description
ThreadedOptimizable(Optimizable.ByCombiningBatchGradient optimizable, InstanceList trainingSet, int numFactors, CacheStaleIndicator cacheIndicator)	Initializes the optimizable and starts new threads.

Method Summary

Modifier and Type	Method	Description
protected void	createTasks()	Creates tasks to be executed in parallel, each task looks at a batch of data.
int	getNumParameters()
Optimizable.ByCombiningBatchGradient	getOptimizable()
double	getParameter(int index)
void	getParameters(double[] buffer)
double	getValue()
void	getValueGradient(double[] buffer)	Returns the gradient, re-computes if gradient is stale.
void	setParameter(int index, double value)
void	setParameters(double[] buff)
void	shutdown()	Shuts down the executor used to start and run threads to compute values and gradients.

Methods inherited from class java.lang.Object

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

Field Detail

trainingSet

protected InstanceList trainingSet

Data

optimizable

protected Optimizable.ByCombiningBatchGradient optimizable

optimizable to be parallelized

batchCachedValue

protected double[] batchCachedValue

Value obtained from the optimizable for each batch

batchCachedGradient

protected java.util.List<double[]> batchCachedGradient

Gradient obtained from the optimizable for each batch

cacheIndicator

protected CacheStaleIndicator cacheIndicator

SLEEP_TIME

public static final int SLEEP_TIME

See Also:

Constant Field Values

Constructor Detail

ThreadedOptimizable

public ThreadedOptimizable(Optimizable.ByCombiningBatchGradient optimizable,
                           InstanceList trainingSet,
                           int numFactors,
                           CacheStaleIndicator cacheIndicator)

Initializes the optimizable and starts new threads.

Parameters:

optimizable - Optimizable to be parallelized

numFactors - Number of factors in model's parameters, used to initialize the gradient

cacheIndicator - Determines when value/gradient become stale

Method Detail

getOptimizable

public Optimizable.ByCombiningBatchGradient getOptimizable()

shutdown

public void shutdown()

Shuts down the executor used to start and run threads to compute values and gradients.

*Note*: For a clean exit of all the threads, it is recommended to call this method after training finishes.

getValue

public double getValue()

Specified by:

getValue in interface Optimizable.ByGradientValue

getValueGradient

public void getValueGradient(double[] buffer)

Returns the gradient, re-computes if gradient is stale.

*Note*: Assumes that buffer is already initialized.

Specified by:

getValueGradient in interface Optimizable.ByGradientValue

createTasks

protected void createTasks()

Creates tasks to be executed in parallel, each task looks at a batch of data.

getNumParameters

public int getNumParameters()

Specified by:

getNumParameters in interface Optimizable

getParameters

public void getParameters(double[] buffer)

Specified by:

getParameters in interface Optimizable

getParameter

public double getParameter(int index)

Specified by:

getParameter in interface Optimizable

setParameters

public void setParameters(double[] buff)

Specified by:

setParameters in interface Optimizable

setParameter

public void setParameter(int index,
                         double value)

Specified by:

setParameter in interface Optimizable