net.pakl.rl
Class ValueFunctionPerceptron

java.lang.Object
  net.pakl.rl.ValueFunctionPerceptron

All Implemented Interfaces:

java.io.Serializable, ValueFunction

Direct Known Subclasses:

ValueFunctionResidualAlgorithmPerceptron

public class ValueFunctionPerceptron
extends java.lang.Object
implements ValueFunction

See Also:

Serialized Form

Field Summary

boolean	CROSS_PRODUCT
int	hiddenUnits
boolean	INCREMENTAL_UPDATE
double	learningRate
double	momentum
double	multiplyAndAdd
boolean	USE_REAL_VECTORS

Constructor Summary

ValueFunctionPerceptron()

ValueFunctionPerceptron(World w)

Method Summary

protected double	dotProduct(double[] u, double[] v)
double	forNetwork(double value)
double	fromNetwork(double value)
java.lang.String	getName() It is important for value functions to have names for the patchwork approach, so that states can identify to which value function patch they belong by the PolicyExtractor.
net.pakl.neuralnet.Perceptron	getNetwork()
double	getValue(State state) Retrieve the value associated with a state (which may be different for non-stored states depending on the actual class implementing this value function).
protected double[]	intsToDoubles(int[] x)
void	setAllowExpansionOfStateBounds(boolean trueOrFalse)
void	setHiddenUnits(int newValue)
void	setLearningRate(double newValue)
void	setMaxMinValue(double newValue)
void	setMomentum(double momentum)
void	setName(java.lang.String name)
void	setRandomSeed(long x)
void	setResidualWeighting(double newValue)
void	setValue(State state, double newValue)
void	setValueOfOutOfBoundsStates(double newValueOfOutOfBoundsStates)
void	setValueOfTerminalStates(double newValue) Many problems define a value of 0 for their terminal states, where the agent can remain.
void	setWorld(World w)
void	storeWeightChangesIfNonIncremental()
protected double	sum(double[] v)
protected double	sum(double[][] v)
java.lang.String	toText()
protected double[][]	vadd(double[][] v1, double[][] v2)
protected double[]	vadd(double[] v1, double[] v2)
protected double[]	vmul(double scalar, double[] vector)
protected double[][]	vmul(double scalar, double[][] vector)
protected double[][]	vsub(double[][] v1, double[][] v2)
protected double[]	vsub(double[] v1, double[] v2)

Methods inherited from class java.lang.Object

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

Field Detail

hiddenUnits

public int hiddenUnits

learningRate

public double learningRate

momentum

public double momentum

multiplyAndAdd

public double multiplyAndAdd

INCREMENTAL_UPDATE

public boolean INCREMENTAL_UPDATE

CROSS_PRODUCT

public boolean CROSS_PRODUCT

USE_REAL_VECTORS

public boolean USE_REAL_VECTORS

Constructor Detail

ValueFunctionPerceptron

public ValueFunctionPerceptron(World w)

ValueFunctionPerceptron

public ValueFunctionPerceptron()

Method Detail

setName

public void setName(java.lang.String name)

Specified by:

setName in interface ValueFunction

getName

public java.lang.String getName()

Description copied from interface: ValueFunction

It is important for value functions to have names for the patchwork approach, so that states can identify to which value function patch they belong by the PolicyExtractor.

Specified by:

getName in interface ValueFunction

setValueOfOutOfBoundsStates

public void setValueOfOutOfBoundsStates(double newValueOfOutOfBoundsStates)

setAllowExpansionOfStateBounds

public void setAllowExpansionOfStateBounds(boolean trueOrFalse)

setWorld

public void setWorld(World w)

Specified by:

setWorld in interface ValueFunction

storeWeightChangesIfNonIncremental

public void storeWeightChangesIfNonIncremental()

getNetwork

public net.pakl.neuralnet.Perceptron getNetwork()

getValue

public double getValue(State state)

Description copied from interface: ValueFunction

Retrieve the value associated with a state (which may be different for non-stored states depending on the actual class implementing this value function).

Specified by:

getValue in interface ValueFunction

sum

protected double sum(double[] v)

sum

protected double sum(double[][] v)

vsub

protected double[] vsub(double[] v1,
                        double[] v2)

vsub

protected double[][] vsub(double[][] v1,
                          double[][] v2)

vadd

protected double[] vadd(double[] v1,
                        double[] v2)

intsToDoubles

protected double[] intsToDoubles(int[] x)

vadd

protected double[][] vadd(double[][] v1,
                          double[][] v2)

vmul

protected double[] vmul(double scalar,
                        double[] vector)

vmul

protected double[][] vmul(double scalar,
                          double[][] vector)

dotProduct

protected double dotProduct(double[] u,
                            double[] v)

setValue

public void setValue(State state,
                     double newValue)

Specified by:

setValue in interface ValueFunction

setResidualWeighting

public void setResidualWeighting(double newValue)

setHiddenUnits

public void setHiddenUnits(int newValue)

setLearningRate

public void setLearningRate(double newValue)

setMomentum

public void setMomentum(double momentum)

setMaxMinValue

public void setMaxMinValue(double newValue)

toText

public java.lang.String toText()

setValueOfTerminalStates

public void setValueOfTerminalStates(double newValue)

Description copied from interface: ValueFunction

Many problems define a value of 0 for their terminal states, where the agent can remain.

Specified by:

setValueOfTerminalStates in interface ValueFunction

forNetwork

public double forNetwork(double value)

fromNetwork

public double fromNetwork(double value)

setRandomSeed

public void setRandomSeed(long x)