burlap.behavior.statehashing

Class DiscretizingStateHashFactory

java.lang.Object

burlap.behavior.statehashing.DiscretizingStateHashFactory

All Implemented Interfaces:

StateHashFactory

public class DiscretizingStateHashFactory
extends java.lang.Object
implements StateHashFactory

This hashing factory is used for comparing states with continuous attributes as if they were discretized, thereby allowing discrete state planning/learning algorithms to be used on domains with continuous attributes. Note that the domain may also include discrete attributes, which will be treated normally in state comparisons.

By default, all continuous attributes will be discretized to their corresponding integer value when making state comparisons and performing hashing. However, attributes may also be floored to some other multiple specified by the client. For instnace, if 0.5 is set as the multiple, then, values will be discretized to 0.5 intervals; e.g., 0.49 would become 0, 0.51 would become 0.5, 9.73 would become 9.5, and 102.3 would become 102. Note using a multiple value of 1.0 is equivlent to floor values to their corresponding int value. Large multiple values result in course discretization and small multiple values result in a fine discretization.

The multiple used can be specified for individual attributes so that different attributes have different degrees of discretization. To set multiple used for an specific attribute, use the addFloorDiscretizingMultipleFor(String, double) method. When a continuous attribute is to be hashed or compared, it is first checked if there has been a specific multiple value set for it. If so, that multiple is used for discretization. If not, the default multiple is used. By defualt, the default multiple is 1.0 (integer floor discretization), but the default multiple may also be set using the setDefaultFloorDiscretizingMultiple(double) method.

Author:

James MacGlashan

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	DiscretizingStateHashFactory.DiscretizedStateHashTuple The StateHashTuple that is generated by objects of the DiscretizingStateHashFactory class.

Field Summary

Fields

Modifier and Type	Field and Description
protected java.util.Map<java.lang.String,java.lang.Double>	attributeWiseMultiples The multiples to use for specific attributes
protected double	defaultMultiple The default multiple to use for any continuous attributes that have not been specifically set.

Constructor Summary

Constructors

Constructor and Description
DiscretizingStateHashFactory() Initializes.
DiscretizingStateHashFactory(double defaultMultiple) Initializes with a specified default multiple to use for discretization.

Method Summary

Methods

Modifier and Type	Method and Description
void	addFloorDiscretizingMultipleFor(java.lang.String attributeName, double nearestMultipleValue) Sets the multiple to use for discretization for the attribute with the specified name.
StateHashTuple	hashState(State s)
protected static int	intMultiples(double mult, double num) Returns int result of num / mult; that is, (int)(num / mult).
void	setDefaultFloorDiscretizingMultiple(double defaultMultiple) Sets the default multiple to use for continuous attributes that do not have specific multiples set for them.

Methods inherited from class java.lang.Object

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

Field Detail

attributeWiseMultiples

protected java.util.Map<java.lang.String,java.lang.Double> attributeWiseMultiples

The multiples to use for specific attributes

defaultMultiple

protected double defaultMultiple

The default multiple to use for any continuous attributes that have not been specifically set. The default is 1.0, whihc corresponds to integer floor discretization.

Constructor Detail

DiscretizingStateHashFactory

public DiscretizingStateHashFactory()

Initializes. the default multiple is set to 1.0 which corresponds to integer floor discretization.

DiscretizingStateHashFactory

public DiscretizingStateHashFactory(double defaultMultiple)

Initializes with a specified default multiple to use for discretization. Any continuous attributes that do not have a specific multiple set for them will use the default.

Parameters:

defaultMultiple - the default multiple to be used by discretization.

Method Detail

addFloorDiscretizingMultipleFor

public void addFloorDiscretizingMultipleFor(java.lang.String attributeName,
                                   double nearestMultipleValue)

Sets the multiple to use for discretization for the attribute with the specified name. See the documentation of this class for more information on how the multiple works. In short, continuous values will be floored to the greatest value that is a multiple of the multiple given and less than or equal to the true value.

Parameters:

attributeName - the name of the attribute whose discretization multiple is being set.

nearestMultipleValue - the multiple to which values are floored.

setDefaultFloorDiscretizingMultiple

public void setDefaultFloorDiscretizingMultiple(double defaultMultiple)

Sets the default multiple to use for continuous attributes that do not have specific multiples set for them. See the documentation of this class for more information on how the multiple works. In short, continuous values will be floored to the greatest value that is a multiple of the multiple given and less than or equal to the true value.

Parameters:

defaultMultiple - the default multiple to which values are floored

hashState

public StateHashTuple hashState(State s)

Specified by:

hashState in interface StateHashFactory

intMultiples

protected static int intMultiples(double mult,
               double num)

Returns int result of num / mult; that is, (int)(num / mult).

Parameters:

mult - the multiple

num - the number

Returns:

the int result of num / mult