edu.udo.cs.yale.generator
Class FeatureGenerator

java.lang.Object
  edu.udo.cs.yale.generator.FeatureGenerator

Direct Known Subclasses:

BinaryNumericalGenerator, ConstantGenerator, SingularNumericalGenerator

public abstract class FeatureGenerator
extends java.lang.Object

Abstract superclass of all attribute generators. Implementing classes have to implement the generate(Example), method and specify the input and output attributes by the appropriate methods so that the using algorithms can use them correctly.

Version:

$Id: FeatureGenerator.java,v 2.30 2006/03/27 13:21:58 ingomierswa Exp $

Author:

Simon Fischer, Ingo Mierswa

Field Summary

private Attribute[]	arguments The argument attributes on which to operate with respect to the example tables attribute array.
private ExampleTable	exampleTable The example table to work on.
private static java.lang.String[]	FUNCTION_NAMES
private static java.lang.Class[]	GENERATOR_CLASSES The classes which corresponds to FUNCTION_NAMES.
private static java.util.Map<java.lang.String,java.lang.Class>	generatorMap Maps function names to generators.
protected Attribute[]	resultAttributes The attributes of the function(s) calculated by this FeatureGenerator.
static int	SELECTION_MODE_ALL Indicates a non-restrictive generator selection mode.
static int	SELECTION_MODE_RESTRICTIVE Indicates a restrictive generator selection mode.
private static int	selectionMode Indicates the selection mode.

Constructor Summary

FeatureGenerator()

Method Summary

private boolean	argumentsOk(ExampleTable input) Checks if the arguments are compatible with the attributes specified by getInputAttributes().
boolean	argumentsSet() returns true, if the arguments have already been set, and false otherwise.
protected boolean	checkCompatibility(Attribute attribute, Attribute compatible, int maxDepth, java.lang.String[] functions)
static FeatureGenerator	createGeneratorForFunction(java.lang.String functionName) Creates a new FeatureGenerator for a given function name.
boolean	equals(java.lang.Object o) A FeatureGenerator equals another FeatureGenerator if its class is equal and its arguments are equal and its function names are equal.
abstract void	generate(DataRow data) Generates the new attribute values for the example e and returns the new attribute values as doubles.
static java.util.List<Attribute>	generateAll(ExampleTable exampleTable, java.util.Collection<FeatureGenerator> generatorList) Generates all new attributes and updates the ExampleTable.
Attribute	getArgument(int i) Returns the i-th selected argument (the true index as used in the example set's example table).
protected ExampleTable	getExampleTable() Gets the example table the examples are from.
abstract java.lang.String	getFunction() Sets the function name.
abstract Attribute[]	getInputAttributes() Returns an array of Attributes where the length is the arity of the generator, [i] is the attribute type of the i-th argument.
abstract java.util.List<Attribute[]>	getInputCandidates(ExampleSet exampleSet, int maxDepth, java.lang.String[] functions) Returns all compatible input attribute arrays for this generator from the given example set as list.
abstract Attribute[]	getOutputAttributes(ExampleTable input) Returns the generated attributes types.
static int	getSelectionMode()
private static java.util.List<Attribute>	newAttributes(FeatureGenerator[] generators, ExampleTable exampleTable) Returns a list of new Attributes that are generated by the given generators.
abstract FeatureGenerator	newInstance() Subclasses must implement this method so that a new instance of this generator class is returned.
static FeatureGenerator	selectGenerator(ExampleSet exampleSet, java.util.List generators, int maxDepth, java.lang.String[] functions, RandomGenerator random) Randomly selects a generator from the generator list.
void	setArguments(Attribute[] args) Sets the arguments (indices) used in future generate(...)
protected void	setExampleTable(ExampleTable et)
abstract void	setFunction(java.lang.String name) Sets the function name.
static void	setSelectionMode(int mode)
java.lang.String	toString()

Methods inherited from class java.lang.Object

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

Field Detail

FUNCTION_NAMES

private static final java.lang.String[] FUNCTION_NAMES

GENERATOR_CLASSES

private static final java.lang.Class[] GENERATOR_CLASSES

The classes which corresponds to FUNCTION_NAMES.

generatorMap

private static java.util.Map<java.lang.String,java.lang.Class> generatorMap

Maps function names to generators.

SELECTION_MODE_ALL

public static int SELECTION_MODE_ALL

Indicates a non-restrictive generator selection mode.

SELECTION_MODE_RESTRICTIVE

public static int SELECTION_MODE_RESTRICTIVE

Indicates a restrictive generator selection mode.

selectionMode

private static int selectionMode

Indicates the selection mode. One of SELECTION_MODE_ALL and SELECTION_MODE_RESTRICTIVE.

resultAttributes

protected Attribute[] resultAttributes

The attributes of the function(s) calculated by this FeatureGenerator.

arguments

private Attribute[] arguments

The argument attributes on which to operate with respect to the example tables attribute array.

exampleTable

private ExampleTable exampleTable

The example table to work on.

Constructor Detail

FeatureGenerator

public FeatureGenerator()

Method Detail

generate

public abstract void generate(DataRow data)
                       throws GenerationException

Generates the new attribute values for the example e and returns the new attribute values as doubles. e.getAttribute(getArgument(i)) is the correct way to access argument i. If the according attribute's type is VALUE_SERIES, the end index can be determined by i_end = getExampleTable().getBlockEndIndex(getArgument(i)). Thus all values of the series can be accessed using indices i through i_end.

Throws:

GenerationException

getInputAttributes

public abstract Attribute[] getInputAttributes()

Returns an array of Attributes where the length is the arity of the generator, [i] is the attribute type of the i-th argument.

getOutputAttributes

public abstract Attribute[] getOutputAttributes(ExampleTable input)

Returns the generated attributes types.

newInstance

public abstract FeatureGenerator newInstance()

Subclasses must implement this method so that a new instance of this generator class is returned. The arguments and the example table will not be cloned and thus be null. This kind of clone is needed as generating algorithms must be able to clone generators form their pool without changing the arguments already set for the others.

setFunction

public abstract void setFunction(java.lang.String name)

Sets the function name. This method is only useful if subclasses can generate more than one function. (like the BasicArithmeticOperationGenerator).

getFunction

public abstract java.lang.String getFunction()

Sets the function name. This method is only useful if subclasses can generate more than one function. (like the BasicArithmeticOperationGenerator).

getInputCandidates

public abstract java.util.List<Attribute[]> getInputCandidates(ExampleSet exampleSet,
                                                               int maxDepth,
                                                               java.lang.String[] functions)

Returns all compatible input attribute arrays for this generator from the given example set as list. Features with a depth greater than maxDepth or which contains one of the given functions should not be used as input candidates. Subclasses must consider if the generator is self-applicable or commutative. A maxDepth of -1 means that no maximal depth should be considered.

checkCompatibility

protected boolean checkCompatibility(Attribute attribute,
                                     Attribute compatible,
                                     int maxDepth,
                                     java.lang.String[] functions)

setExampleTable

protected void setExampleTable(ExampleTable et)

getExampleTable

protected ExampleTable getExampleTable()

Gets the example table the examples are from.

setArguments

public void setArguments(Attribute[] args)

Sets the arguments (indices) used in future generate(...) calls and has to be called prior to any generate(...) calls. The caller of this method has to take care that:

• args.length == getInputAttributes().length, i.e. that the arity is correct.
• The types of the example attributes match the types specified by getInputAttributes().
• The true attribute indices are used (as used by the example set's example table)

argumentsSet

public boolean argumentsSet()

returns true, if the arguments have already been set, and false otherwise.

getArgument

public Attribute getArgument(int i)

Returns the i-th selected argument (the true index as used in the example set's example table).

argumentsOk

private boolean argumentsOk(ExampleTable input)

Checks if the arguments are compatible with the attributes specified by getInputAttributes().

createGeneratorForFunction

public static FeatureGenerator createGeneratorForFunction(java.lang.String functionName)

Creates a new FeatureGenerator for a given function name.

selectGenerator

public static FeatureGenerator selectGenerator(ExampleSet exampleSet,
                                               java.util.List generators,
                                               int maxDepth,
                                               java.lang.String[] functions,
                                               RandomGenerator random)

Randomly selects a generator from the generator list. The probability of a generator to be selected is proportional to its number of attribute combinations as delivered by getInputCandidates(ExampleSet, int, String[]) method. Returns null if no generators are applicable.

Parameters:

generators - List of FeatureGenerators

generateAll

public static java.util.List<Attribute> generateAll(ExampleTable exampleTable,
                                                    java.util.Collection<FeatureGenerator> generatorList)
                                             throws GenerationException

Generates all new attributes and updates the ExampleTable. Returns a list of Attributes for the newly generated attributes.

Parameters:

exampleTable - the source example table

generatorList - List of FeatureGenerators

Returns:

A list of Attributes

Throws:

GenerationException

newAttributes

private static java.util.List<Attribute> newAttributes(FeatureGenerator[] generators,
                                                       ExampleTable exampleTable)

Returns a list of new Attributes that are generated by the given generators.

getSelectionMode

public static int getSelectionMode()

setSelectionMode

public static void setSelectionMode(int mode)

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object

equals

public boolean equals(java.lang.Object o)

A FeatureGenerator equals another FeatureGenerator if its class is equal and its arguments are equal and its function names are equal.

Overrides:

equals in class java.lang.Object