edu.iu.pcl.absurdist.absurdist
Class Absurdist

java.lang.Object
  |
  +--edu.iu.pcl.absurdist.absurdist.Absurdist

public class Absurdist

extends java.lang.Object

Class Absurdist provides the ABSURDIST II graph matching algorithms to compute the correspondence between two concept systems, and finds the optimal mappings between them. The behavior of this class is affected by a number of parameters, which can be supplied as Java System Properties on the command line (java -Dname=value ...). For example, running java -DresetL=false -DL=0.01 -DmaxLoop=5000 driver.ApplicDriver will run the program with fixed L equal to 0.01, interrupting the loop at the 5000-th iteration, if the convergence has not been achieved. java -DresetL=true driver.ApplicDriver will adjust the learning rate at every step to 1/2 of the maximum safe rate. This is supposed to achieve fastest convergence. One can find all the parameters in the code by searching for 'prop.getOption'.

Nested Class Summary

static class

Absurdist.Results Used for reporting.

Field Summary

static boolean	adjBest If true, we'll find the optimal permutation (this may be very expensive!)
double	alpha Coefficients for external similarity.
static double	alphaShare This flag determines the alpha/(alpha+beta) ratio to be used in adjustCoeff(); it only is used if useAlphaShare is true, and is ignored otherwise.
double	beta Coefficients for activation.
double	chi Coefficients for inhibition.
static double	cInit Intitial value for correspondence matrix.
static int	coefMode This parameter controls how chi and beta are initially set, and whether they are reset during the iterative process.
static int	CONVERGED Concept systems mapping iteration converged.
double[][][]	correspond Correspondence matrix C[step][row][col].
static int	CREATED Concept system created.
static double	cvgThresh Threshold for iteration convergence.
static int	dampingMode Damping mode.
static boolean	enforce1to1map If true, 1-1 mapping will be enforced.
double[][]	exSim External similarity E[row][col].
double[][][]	inhib Inhibitor I[step][row][col].
static int	INITIAL Initial status without concept system created.
double[][][]	inSim Internal similarity, also called activator, R[step][row][col].
static int	intLoop Sampling interval for plots.
static int	ITERATING Concept systems in mapping iteration.
static double	lRate Learning rate (this may be adjusted if resetL=true).
int[]	mapNO Mappings from systemN to systemO, recording indices of matched concepts.
int[]	mapON Mappings from systemO to systemN, recording indices of matched concepts.
static int	MAPPED Concept systems mapping completed.
static double	mapThresh Threshold for valid mapping.
static double	maxL The ceiling for adjustable learning rate.
static int	maxLoop Maximum number of iterations.
double[][][]	netIn Network input N[step][row][col].
int	nStep Actual number of steps being recorded during network iteration.
static ParseConfig	prop An empty property list -- just an interface to system properties
static boolean	resetL If true, we recompute the learning rate at each step, for fastest convergence.
Absurdist.Results	results
int	status Status of the system.
ConceptSystem	systemN Target system for mapping.
ConceptSystem	systemO Source system for mapping.
boolean	systemsAreSimple This will be set to true in the constructor if both concept systems being matched are "binary", that is have no other relationship weights than 1 and 0.
boolean	useAlphaShare If true, the alphaShare parameter is used to set alpha via beta.
boolean	useAnchors If true, use anchors overlap for external similarity
static boolean	useExponential If true, use the exponential version of activator and inhibitor to compute netwoek input.
static boolean	useFastGeneral If true, use the "fast" (N^2 D^2) method for general (labeled, weighted) graphs.

Constructor Summary

Absurdist()
Creates an empty absurdist system.

Absurdist(ConceptSystem systemO, ConceptSystem systemN)
Creates an absurdist system containing the source and target concept systems and random external similarities.

Method Summary

double	calcLinkSim(Link lo, Link ln) Computes the similarity between the two links.
void	evalMap() Evaluates the quality of the mapping in mapON[].
void	initExSim() Initialize external similarity matrix to all 0's.
boolean	isConverged() Returns true if the network iteration is converged; otherwise false.
boolean	isCreated() Returns true if the two systems are already created; otherwise false.
boolean	isInitial() Returns true if the two systems have not yet been created; otherwise false.
boolean	isIterating() Returns true if the network iteration is going on; otherwise false.
boolean	isMapped() Returns true if the two systems are already mapped; otherwise false.
static Absurdist	loadSystems(java.lang.String fileName, int w, int h) Creates a new instance of Absurdist initialized with the 2 systems read from a file containing Java serialized objects.
void	mapSystem() Maps orginal concept system to the one with added noise, using exponenital/linear activator/inhibitor.
void	print() Prints the source and target concept systems, the correspondence matrix, and the final mappings.
void	printCorrespond() Prints the correspondence matrix to the standard output.
void	printExSim() Prints the external similarity matrix to the standard output.
void	printMap() Prints the mapping between systemO and systemN to the standard output.
void	printSystems() Prints the two systems to the standard output.
void	readjustChi(double[][] exSim, double[][] inSim, double[][] inhib) A normalization method that takes both exSim and inSim into account.
void	saveSystems(java.lang.String fileName) Writes out the concept systems as Java serialized objects intoa file.
void	setSystemN(ConceptSystem systemN) Sets the target concepts system to be mapped to.
void	setSystemO(ConceptSystem systemO) Sets the source concepts system to be mapped from.
void	setSystems(ConceptSystem systemO, ConceptSystem systemN) Sets the two concepts systems for mapping.
double	sumElem(double[][] a) Computes the sum of the elements in an array.
static void	turnOffHistory() Turn off the iteration history recording so that less memory will be used.
void	writeCorrespond(java.lang.String filename) Writes the correspondence matrix to a text file.
void	writeMap(java.lang.String filename) Writes the mapping between systemO and systemN to a text file.

Methods inherited from class java.lang.Object

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

Field Detail

prop

public static ParseConfig prop

An empty property list -- just an interface to system properties

INITIAL

public static final int INITIAL

Initial status without concept system created.

See Also:

Constant Field Values

CREATED

public static final int CREATED

Concept system created.

See Also:

Constant Field Values

ITERATING

public static final int ITERATING

Concept systems in mapping iteration.

See Also:

Constant Field Values

CONVERGED

public static final int CONVERGED

Concept systems mapping iteration converged.

See Also:

Constant Field Values

MAPPED

public static final int MAPPED

Concept systems mapping completed.

See Also:

Constant Field Values

cInit

public static double cInit

Intitial value for correspondence matrix.

lRate

public static double lRate

Learning rate (this may be adjusted if resetL=true).

maxLoop

public static int maxLoop

Maximum number of iterations.

intLoop

public static int intLoop

Sampling interval for plots.

cvgThresh

public static double cvgThresh

Threshold for iteration convergence.

mapThresh

public static double mapThresh

Threshold for valid mapping.

coefMode

public static int coefMode

This parameter controls how chi and beta are initially set, and whether they are reset during the iterative process.

resetL

public static boolean resetL

If true, we recompute the learning rate at each step, for fastest convergence. (For compatibility with the 2003 code, set the default to false!)

maxL

public static double maxL

The ceiling for adjustable learning rate. If 0 or negative, it is ignored

dampingMode

public static int dampingMode

Damping mode. Should be 0 (= G & R 2002) or 1 (= Hard damping)

useFastGeneral

public static boolean useFastGeneral

If true, use the "fast" (N^2 D^2) method for general (labeled, weighted) graphs.

useExponential

public static boolean useExponential

If true, use the exponential version of activator and inhibitor to compute netwoek input.

enforce1to1map

public static boolean enforce1to1map

If true, 1-1 mapping will be enforced.

adjBest

public static boolean adjBest

If true, we'll find the optimal permutation (this may be very expensive!) and compare the quality of our permutation to it

systemsAreSimple

public boolean systemsAreSimple

This will be set to true in the constructor if both concept systems being matched are "binary", that is have no other relationship weights than 1 and 0. If this is the case, we can use faster algorithms for computing inhibition, excitation, etc.

useAnchors

public boolean useAnchors

If true, use anchors overlap for external similarity

useAlphaShare

public boolean useAlphaShare

If true, the alphaShare parameter is used to set alpha via beta. This is false by default.

alphaShare

public static double alphaShare

This flag determines the alpha/(alpha+beta) ratio to be used in adjustCoeff(); it only is used if useAlphaShare is true, and is ignored otherwise. If it is 1.0, then alpha will be set to 1.0 and beta to 0.

See Also:

#adjustCoeff

alpha

public double alpha

Coefficients for external similarity.

beta

public double beta

Coefficients for activation.

chi

public double chi

Coefficients for inhibition.

systemO

public ConceptSystem systemO

Source system for mapping.

systemN

public ConceptSystem systemN

Target system for mapping.

nStep

public int nStep

Actual number of steps being recorded during network iteration.

status

public int status

Status of the system.

exSim

public double[][] exSim

External similarity E[row][col].

inSim

public double[][][] inSim

Internal similarity, also called activator, R[step][row][col].

inhib

public double[][][] inhib

Inhibitor I[step][row][col].

netIn

public double[][][] netIn

Network input N[step][row][col].

correspond

public double[][][] correspond

Correspondence matrix C[step][row][col].

mapON

public int[] mapON

Mappings from systemO to systemN, recording indices of matched concepts.

mapNO

public int[] mapNO

Mappings from systemN to systemO, recording indices of matched concepts.

results

public Absurdist.Results results

Constructor Detail

Absurdist

public Absurdist()

Creates an empty absurdist system.

Absurdist

public Absurdist(ConceptSystem systemO,
                 ConceptSystem systemN)

Creates an absurdist system containing the source and target concept systems and random external similarities.

Method Detail

setSystemO

public void setSystemO(ConceptSystem systemO)

Sets the source concepts system to be mapped from.

setSystemN

public void setSystemN(ConceptSystem systemN)

Sets the target concepts system to be mapped to.

setSystems

public void setSystems(ConceptSystem systemO,
                       ConceptSystem systemN)

Sets the two concepts systems for mapping.

initExSim

public void initExSim()

Initialize external similarity matrix to all 0's.

turnOffHistory

public static void turnOffHistory()

Turn off the iteration history recording so that less memory will be used. This is used if you don't need to keep the intermediate results during the iteration. This is important for saving memory, and should always be done when you don't use the GUI and don't need the "timeseries" graphs.

calcLinkSim

public double calcLinkSim(Link lo,
                          Link ln)

Computes the similarity between the two links. Each link may be a "real" link (a pair of numbers), or the EMPTY link.

readjustChi

public void readjustChi(double[][] exSim,
                        double[][] inSim,
                        double[][] inhib)

A normalization method that takes both exSim and inSim into account. The goal is to have sum_{a,b}( (alpha*exSim + beta*inSim - chi*inhib)[a][b]) = 0. Learning rate may be recomputed too.

sumElem

public double sumElem(double[][] a)

Computes the sum of the elements in an array.

evalMap

public void evalMap()

Evaluates the quality of the mapping in mapON[].

Note: mismapped concepts and relations only apply to mappings between a concept system and its noisy copy, assuming that the noise is small enough so that the mapping should be identity, i.e. Ci -> Ci'.

mapSystem

public void mapSystem()

Maps orginal concept system to the one with added noise, using exponenital/linear activator/inhibitor. systemO and systemN have to be set preceding to calling this method.

isInitial

public boolean isInitial()

Returns true if the two systems have not yet been created; otherwise false.

isCreated

public boolean isCreated()

Returns true if the two systems are already created; otherwise false.

isIterating

public boolean isIterating()

Returns true if the network iteration is going on; otherwise false.

isConverged

public boolean isConverged()

Returns true if the network iteration is converged; otherwise false.

isMapped

public boolean isMapped()

Returns true if the two systems are already mapped; otherwise false.

writeCorrespond

public void writeCorrespond(java.lang.String filename)
                     throws java.lang.Exception

Writes the correspondence matrix to a text file.

java.lang.Exception

writeMap

public void writeMap(java.lang.String filename)
              throws java.lang.Exception

Writes the mapping between systemO and systemN to a text file.

java.lang.Exception

printSystems

public void printSystems()

Prints the two systems to the standard output.

printExSim

public void printExSim()

Prints the external similarity matrix to the standard output.

printCorrespond

public void printCorrespond()

Prints the correspondence matrix to the standard output.

printMap

public void printMap()

Prints the mapping between systemO and systemN to the standard output.

print

public void print()
           throws java.lang.Exception

Prints the source and target concept systems, the correspondence matrix, and the final mappings.

java.lang.Exception

loadSystems

public static Absurdist loadSystems(java.lang.String fileName,
                                    int w,
                                    int h)

Creates a new instance of Absurdist initialized with the 2 systems read from a file containing Java serialized objects. Such a file may have been produced with saveSystems(). For display purposes, both graph are to be fitted into a w-by-x panel, splitting it into 2 halves.

Parameters:

w - Width of the display area for the graphs.

h - Height of the display area for the graphs.

saveSystems

public void saveSystems(java.lang.String fileName)

Writes out the concept systems as Java serialized objects intoa file. This method can be used if you want to restore the data later from the file, using loadSystems().