edu.stanford.nlp.optimization

Class SparseAdaGradMinimizer<K,F extends SparseOnlineFunction<K>>

java.lang.Object

edu.stanford.nlp.optimization.SparseAdaGradMinimizer<K,F>

All Implemented Interfaces:

SparseMinimizer<K,F>

public class SparseAdaGradMinimizer<K,F extends SparseOnlineFunction<K>>
extends java.lang.Object
implements SparseMinimizer<K,F>

AdaGrad optimizer that works online, and use sparse gradients, need a function that takes a Counter<K> as argument and returns a Counter<K> as gradient

Author:

Sida Wang

Field Summary

Fields

Modifier and Type	Field and Description
protected int	batchSize
double	EPS
protected double	eta
protected double	lambdaL1
protected double	lambdaL2
protected int	numPasses
boolean	quiet
protected java.util.Random	randGenerator
double	soften
protected Counter<K>	sumGradSquare
protected Counter<K>	x

Constructor Summary

Constructors

Constructor and Description
SparseAdaGradMinimizer(int numPasses)
SparseAdaGradMinimizer(int numPasses, double eta)
SparseAdaGradMinimizer(int numPasses, double eta, int batchSize, double lambdaL1, double lambdaL2)

Method Summary

Modifier and Type	Method and Description
protected java.lang.String	getName()
Counter<K>	minimize(F function, Counter<K> initial) Attempts to find an unconstrained minimum of the objective function starting at initial, within functionTolerance.
Counter<K>	minimize(F function, Counter<K> x, int maxIterations)
protected void	sayln(java.lang.String s)

Methods inherited from class java.lang.Object

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

Field Detail

quiet

public boolean quiet

numPasses

protected int numPasses

batchSize

protected int batchSize

eta

protected double eta

lambdaL1

protected double lambdaL1

lambdaL2

protected double lambdaL2

sumGradSquare

protected Counter<K> sumGradSquare

x

protected Counter<K> x

randGenerator

protected java.util.Random randGenerator

EPS

public final double EPS

See Also:

Constant Field Values

soften

public final double soften

See Also:

Constant Field Values

Constructor Detail

SparseAdaGradMinimizer

public SparseAdaGradMinimizer(int numPasses)

SparseAdaGradMinimizer

public SparseAdaGradMinimizer(int numPasses,
                              double eta)

SparseAdaGradMinimizer

public SparseAdaGradMinimizer(int numPasses,
                              double eta,
                              int batchSize,
                              double lambdaL1,
                              double lambdaL2)

Method Detail

minimize

public Counter<K> minimize(F function,
                           Counter<K> initial)

Description copied from interface: SparseMinimizer

Attempts to find an unconstrained minimum of the objective function starting at initial, within functionTolerance.

Specified by:

minimize in interface SparseMinimizer<K,F extends SparseOnlineFunction<K>>

Parameters:

function - the objective function

initial - a initial feasible point

Returns:

Unconstrained minimum of function

minimize

public Counter<K> minimize(F function,
                           Counter<K> x,
                           int maxIterations)

Specified by:

minimize in interface SparseMinimizer<K,F extends SparseOnlineFunction<K>>

getName

protected java.lang.String getName()

sayln

protected void sayln(java.lang.String s)