org.apache.ignite.ml.optimization.updatecalculators

Class NesterovUpdateCalculator<M extends SmoothParametrized<M>>

java.lang.Object

org.apache.ignite.ml.optimization.updatecalculators.NesterovUpdateCalculator<M>

All Implemented Interfaces:

Serializable, ParameterUpdateCalculator<M,NesterovParameterUpdate>

public class NesterovUpdateCalculator<M extends SmoothParametrized<M>>
extends Object
implements ParameterUpdateCalculator<M,NesterovParameterUpdate>

Class encapsulating Nesterov algorithm for MLP parameters updateCache.

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected double	momentum Momentum constant.

Constructor Summary

Constructors

Constructor and Description
NesterovUpdateCalculator(double learningRate, double momentum) Construct NesterovUpdateCalculator.

Method Summary

Modifier and Type	Method and Description
NesterovParameterUpdate	calculateNewUpdate(M mdl, NesterovParameterUpdate updaterParameters, int iteration, Matrix inputs, Matrix groundTruth) Calculate new update.
NesterovParameterUpdate	init(M mdl, IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss) Initializes the update calculator.
<M1 extends M> M1	update(M1 obj, NesterovParameterUpdate update) Update given obj with this parameters.

Methods inherited from class java.lang.Object

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

Field Detail

momentum

protected double momentum

Momentum constant.

Constructor Detail

NesterovUpdateCalculator

public NesterovUpdateCalculator(double learningRate,
                                double momentum)

Construct NesterovUpdateCalculator.

Parameters:

momentum - Momentum constant.

Method Detail

calculateNewUpdate

public NesterovParameterUpdate calculateNewUpdate(M mdl,
                                                  NesterovParameterUpdate updaterParameters,
                                                  int iteration,
                                                  Matrix inputs,
                                                  Matrix groundTruth)

Calculate new update.

Specified by:

calculateNewUpdate in interface ParameterUpdateCalculator<M extends SmoothParametrized<M>,NesterovParameterUpdate>

Parameters:

mdl - Model to be updated.

updaterParameters - Updater parameters to updateCache.

iteration - Current trainer iteration.

inputs - Inputs.

groundTruth - True values.

Returns:

Updated parameters.

init

public NesterovParameterUpdate init(M mdl,
                                    IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss)

Initializes the update calculator.

Specified by:

init in interface ParameterUpdateCalculator<M extends SmoothParametrized<M>,NesterovParameterUpdate>

Parameters:

mdl - Model to be trained.

loss - Loss function.

Returns:

Initialized parameters.

update

public <M1 extends M> M1 update(M1 obj,
                                NesterovParameterUpdate update)

Update given obj with this parameters.

Specified by:

update in interface ParameterUpdateCalculator<M extends SmoothParametrized<M>,NesterovParameterUpdate>

Parameters:

obj - Object to be updated.