de.jungblut.online.minimizer

Class StochasticGradientDescent.StochasticGradientDescentBuilder

java.lang.Object

de.jungblut.online.minimizer.StochasticGradientDescent.StochasticGradientDescentBuilder

Enclosing class:

StochasticGradientDescent

public static class StochasticGradientDescent.StochasticGradientDescentBuilder
extends Object

Method Summary

Modifier and Type	Method and Description
StochasticGradientDescent.StochasticGradientDescentBuilder	breakOnDifference(double delta) Breaks minimization process when the given delta in costs have been archieved.
StochasticGradientDescent	build()
static StochasticGradientDescent.StochasticGradientDescentBuilder	create(double alpha) Creates a new builder.
StochasticGradientDescent.StochasticGradientDescentBuilder	enableAdaptiveLearningRate() Enables adaptive learning rate, using the algorithm: alpha = 1d / (initialAlpha * (allIterations + 2)); where allIterations is a counter over all passes.
StochasticGradientDescent.StochasticGradientDescentBuilder	historySize(int historySize) Sets the size of the history to keep to compute average improvements and output progress information.
StochasticGradientDescent.StochasticGradientDescentBuilder	holdoutValidationPercentage(double perc) Holdout validation percentage, this will take a subset of the data on the stream and do a validation on it.
StochasticGradientDescent.StochasticGradientDescentBuilder	lambda(double lambda) Sets the regularization parameter "lambda".
StochasticGradientDescent.StochasticGradientDescentBuilder	momentum(double momentum) Add momentum to this gradient descent minimizer.
StochasticGradientDescent.StochasticGradientDescentBuilder	progressReportInterval(int interval) Sets the progress report interval.
StochasticGradientDescent.StochasticGradientDescentBuilder	weightUpdater(WeightUpdater weightUpdater) Sets the weight updater, for example to use regularization.

Methods inherited from class java.lang.Object

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

Method Detail

build

public StochasticGradientDescent build()

momentum

public StochasticGradientDescent.StochasticGradientDescentBuilder momentum(double momentum)

Add momentum to this gradient descent minimizer.

Parameters:

momentum - the momentum to use. Between 0 and 1.

Returns:

the builder again.

holdoutValidationPercentage

public StochasticGradientDescent.StochasticGradientDescentBuilder holdoutValidationPercentage(double perc)

Holdout validation percentage, this will take a subset of the data on the stream and do a validation on it.

Parameters:

perc - the percentage to use. Between 0 and 1.

Returns:

the builder again.

weightUpdater

public StochasticGradientDescent.StochasticGradientDescentBuilder weightUpdater(WeightUpdater weightUpdater)

Sets the weight updater, for example to use regularization. The default is the normal gradient descent. To set the regularization parameter use the lambda(double) method.

Parameters:

weightUpdater - the updater to use.

Returns:

the builder again.

lambda

public StochasticGradientDescent.StochasticGradientDescentBuilder lambda(double lambda)

Sets the regularization parameter "lambda".

Parameters:

lambda - the amount to regularize with.

Returns:

the builder again.

historySize

public StochasticGradientDescent.StochasticGradientDescentBuilder historySize(int historySize)

Sets the size of the history to keep to compute average improvements and output progress information.

Returns:

the builder again.

progressReportInterval

public StochasticGradientDescent.StochasticGradientDescentBuilder progressReportInterval(int interval)

Sets the progress report interval. Since writing to the console/log might be expensive, this is an easy way to limit the logging if needed.

Parameters:

interval - the interval. E.g. every 10th iteration.

Returns:

the builder again.

breakOnDifference

public StochasticGradientDescent.StochasticGradientDescentBuilder breakOnDifference(double delta)

Breaks minimization process when the given delta in costs have been archieved. Usually a quite low value of 1e-4 to 1e-8.

Parameters:

delta - the delta to break in difference between two costs.

Returns:

the builder again.

enableAdaptiveLearningRate

public StochasticGradientDescent.StochasticGradientDescentBuilder enableAdaptiveLearningRate()

Enables adaptive learning rate, using the algorithm:

 alpha = 1d / (initialAlpha * (allIterations + 2));
 

where allIterations is a counter over all passes.

Returns:

the builder again

create

public static StochasticGradientDescent.StochasticGradientDescentBuilder create(double alpha)

Creates a new builder.

Parameters:

alpha - the learning rate to set.

Returns:

a new builder.