de.jungblut.online.ml

Class AbstractMinimizingOnlineLearner<M extends Model>

java.lang.Object

de.jungblut.online.ml.AbstractOnlineLearner<M>

de.jungblut.online.ml.AbstractMinimizingOnlineLearner<M>

All Implemented Interfaces:

OnlineLearner<M>

Direct Known Subclasses:

RegressionLearner

public abstract class AbstractMinimizingOnlineLearner<M extends Model>
extends AbstractOnlineLearner<M>

Field Summary

Fields

Modifier and Type	Field and Description
protected StochasticMinimizer	minimizer
protected int	numPasses
protected Random	random
protected boolean	sparseWeights

Fields inherited from class de.jungblut.online.ml.AbstractOnlineLearner

featureDimension, numOutcomeClasses, outcomeDimension, streamSupplier, verbose

Constructor Summary

Constructors

Constructor and Description
AbstractMinimizingOnlineLearner(StochasticMinimizer minimizer)

Method Summary

Modifier and Type	Method and Description
protected abstract M	createModel(de.jungblut.math.DoubleVector weights) Creates a model with the given minimized weights.
protected abstract de.jungblut.math.minimize.CostGradientTuple	observeExample(de.jungblut.online.ml.FeatureOutcomePair next, de.jungblut.math.DoubleVector weights) Observes the next example.
protected de.jungblut.math.minimize.CostGradientTuple	observeExampleSafe(de.jungblut.online.ml.FeatureOutcomePair next, de.jungblut.math.DoubleVector weights)
protected de.jungblut.math.DoubleVector	randomInitialize(int dimension)
void	setNumPasses(int passes)
void	setRandom(Random random)
M	train(java.util.function.Supplier<java.util.stream.Stream<de.jungblut.online.ml.FeatureOutcomePair>> streamSupplier) Trains a new model using the supplied streams.
void	useSparseWeights()

Methods inherited from class de.jungblut.online.ml.AbstractOnlineLearner

init, peekDimensions, setVerbose, verbose

Methods inherited from class java.lang.Object

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

Field Detail

minimizer

protected final StochasticMinimizer minimizer

random

protected Random random

numPasses

protected int numPasses

sparseWeights

protected boolean sparseWeights

Constructor Detail

AbstractMinimizingOnlineLearner

public AbstractMinimizingOnlineLearner(StochasticMinimizer minimizer)

Method Detail

train

public M train(java.util.function.Supplier<java.util.stream.Stream<de.jungblut.online.ml.FeatureOutcomePair>> streamSupplier)

Description copied from interface: OnlineLearner

Trains a new model using the supplied streams. In case an algorithm needs multiple iterations, it simply gets a new one from the supplier.

Parameters:

streamSupplier - the supplier that creates a new stream that can be consumed.

Returns:

the trained model.

observeExample

protected abstract de.jungblut.math.minimize.CostGradientTuple observeExample(de.jungblut.online.ml.FeatureOutcomePair next,
                                                                              de.jungblut.math.DoubleVector weights)

Observes the next example.

Parameters:

next - the next feature/outcome pair.

weights - the current weights.

Returns:

a cost gradient tuple that can be used for minimization.

createModel

protected abstract M createModel(de.jungblut.math.DoubleVector weights)

Creates a model with the given minimized weights.

Parameters:

weights - the learned weights.

Returns:

a model that describes the weights.

observeExampleSafe

protected de.jungblut.math.minimize.CostGradientTuple observeExampleSafe(de.jungblut.online.ml.FeatureOutcomePair next,
                                                                         de.jungblut.math.DoubleVector weights)

randomInitialize

protected de.jungblut.math.DoubleVector randomInitialize(int dimension)

setRandom

public void setRandom(Random random)

useSparseWeights

public void useSparseWeights()

setNumPasses

public void setNumPasses(int passes)