tc
Class TCExperiment

java.lang.Object
  tc.TCExperiment

public class TCExperiment

extends java.lang.Object

This class represents an experiment with one corpus. it holds the properties of the experiment (weighting function, stemming, stopword removal, dimensionality, classifier properties etc.) and provides methods to load and test the corpus, fill tables, create graphs for the GUI, ...

Field Summary

(package private) java.util.ArrayList	batchEvaluationResults holds all the batch evaluations made so far
(package private) TCCorpusLoader	corpusLoader the loader for the corpus
(package private) java.lang.String	corpusName holds the name of the corpus
(package private) TCCorpusDistributionsData	distributionsDataCorpus contains the mean and standard deviation of some gaussian distributions
(package private) TCCorpusDistributionsData	distributionsDataForTermsOfInterest contains the mean and standard deviation of some gaussian distributions
(package private) int	featureSpaceDimension the current feature space dimension (dimension of the feature term vector)
(package private) int	kNN_k number of nearest neighbors used for the knn algorithm
(package private) double	kNNThreshold the threshold for the kNN classifier
(package private) double	naiveBayesThreshold the threshold for the naive bayes classifier
(package private) javax.swing.JFrame	parentFrame the parent frame (used for progress monitoring etc...)
(package private) java.util.ArrayList[]	precisionRecallResults holds all the (recall, precision) values for the precision and recall graph
(package private) int	stemmerMethod indicates which stemming method should be used
(package private) java.lang.String	stopWordsString the string containing all the words considered as stop words
(package private) TCCorpus	testCorpus the actual corpus used for this experiment
(package private) boolean	useStopWordRemoval indicates if stop words should be removed
(package private) int	weightFunctionIndex the index of the weighting function currently used
(package private) int	weightFunctionSumIndex the method of sumation of the weighting function currently in use

Constructor Summary

TCExperiment()
constructor.

Method Summary

java.lang.String	classifyDocumentAndProvideReport(java.lang.String path, int classifierIndex) classifies a single document and provides a textual report
TCAccuracyResult	computeClassifierAccuracyKNN(double stepSize) computes the accuracy (break even, 11pt precision, precision-recall values, etc.) for the kNN classifier.
TCAccuracyResult	computeClassifierAccuracyNBB(double stepSize) computes the accuracy (break even, precision-recall values) for the naive bayes classifier.
boolean	computeWeightValues(int minimumDFTFThreshold, javax.swing.ProgressMonitor progressMonitor) computes the weight values for the experiment corpus
Graph2D	createEvaluationGraph(int evaluationNumber) creates the batch evaluation graph where different setting (e.g.
Graph2D	createGDGraph(int distributionIndex) creates the gaussian distributions graph for later use
Graph2D	createPrecisionRecallGraph(int classifierIndex) creates the precision<->recall graph based on an ArrayList with evaluation results for many different thresholds
TCEvaluationResult	evaluateClassification(int classifierIndex) Evaluates the kNN or NBB classificator with the current settings and calculates recall and precision.
void	evaluateClassificationBatch() Evaluates (precision and/or recall) one or more classifiers for different types of setting (weighting function, properties of the classificatiob algorithm, number of unique terms, etc.) and fills a TCBatchResult object which can be saved to disk or used for plotting the curves...
TCEvaluationResult	evaluateKNNClassificationForThreshold(double threshold, boolean display) Evaluates the kNN classificator with the current settings (and one single threshold) and calculates recall and precision.
TCEvaluationResult	evaluateNBBClassificationForThreshold(double threshold, boolean display) Evaluates the naive bayes bernoulli classificator with the current settings (and one single threshold) and calculates recall and precision.
void	exportDocumentVectors(java.lang.String path, int documentType)
TCTableData	fillBatchEvaluationDataTable() creates the table with the batch evaluation data
TCTableData	fillCategoryDataTable() resembles the data of every category (number of document, etc.)
TCTableData	fillDocumentDataTable(int property) fills the table with properties of all the documents
TCTableData	fillTermDataTable() shows the properties of all the terms in the terms of interest arraylist (which are the terms after the dimensionality reduction step)
int	getNumberOfCategories() returns the number of categories in the experiment corpus
int	getNumberOfDocuments() returns the number of documents in the experiment corpus
int	getNumberOfLabeledDocuments() returns the number of documents which were assigned to at least one category
int	getNumberOfProcessedWords() the number of processed words (the number of words which were sorted into the term vectors by the loadCorpus algorithms
int	getNumberOfTestDocuments() returns the number of test documents currently loaded
int	getNumberOfTestDocumentsWithLabel() returns the number of test documents with at least one label (which are essigned to at least one category)
int	getNumberOfUniqueTerms() returns the number of distinct (unique) terms in the corpus
double	getNumberOfWordsPerDocument() returns the average number of words per document in the corpus
java.lang.String	getStopWordsString()
int	getWeightFunctionIndex() returns the weight function index currently in use
int	getWeightFunctionSumMethodIndex() returns the index of the weight function sumation method
void	loadBatchEvaluationResult(java.lang.String path)
void	loadClassificatorScoring(java.lang.String path, int classifierIndex) loads a scoring file.
void	loadCorpus(boolean useSWR, int stemMethod) loads one of the implemented corpora.
void	loadStopWordsString() loads the file with all the stop words and saves it in a String for later use
double	reduceFeatureSpaceDimension(int fSD, int minimumDFTFFrequency) reduces the feature space and calculates the distributions and he tfidf values for all the documents in the corpus
void	saveBatchEvaluationResult(java.lang.String path, int resultIndex) saves a batch evaluation to the given path
void	saveClassificatorScoring(java.lang.String path, int classifierIndex) saves the classificator scorings for later use.
void	setCorpusName(java.lang.String name) sets the label of the corpus
void	setKNNk(int k) sets the k value for the kNN classifier
void	setKNNThreshold(double t) sets the threshold for the naive bayes classifier
void	setNaiveBayesThreshold(double t) sets the threshold for the naive bayes classifier
void	setParentJFrame(javax.swing.JFrame frame) sets the JFrame this class was created from (used for progress monitoring etc.)
void	setWeightFunctionIndex(int wfindex) sets the weighting function which should be used for the dimensionality reduction step
void	setWeightFunctionSumMethodIndex(int wfsmi) sets the method of sumation for the weighting function currently in use
void	trainKNNClassifier() calculates all the kNN similarity scores for every test documents loaded by the corpusLoader.
void	trainNaiveBayesClassifier() trains the naive bayes classifier (calculates the category probabilities for every document) results are saved in the TCCorpus class for later processing

Methods inherited from class java.lang.Object

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

Field Detail

corpusName

java.lang.String corpusName

holds the name of the corpus

stemmerMethod

int stemmerMethod

indicates which stemming method should be used

useStopWordRemoval

boolean useStopWordRemoval

indicates if stop words should be removed

weightFunctionIndex

int weightFunctionIndex

the index of the weighting function currently used

weightFunctionSumIndex

int weightFunctionSumIndex

the method of sumation of the weighting function currently in use

featureSpaceDimension

int featureSpaceDimension

the current feature space dimension (dimension of the feature term vector)

kNN_k

int kNN_k

number of nearest neighbors used for the knn algorithm

naiveBayesThreshold

double naiveBayesThreshold

the threshold for the naive bayes classifier

kNNThreshold

double kNNThreshold

the threshold for the kNN classifier

stopWordsString

java.lang.String stopWordsString

the string containing all the words considered as stop words

testCorpus

TCCorpus testCorpus

the actual corpus used for this experiment

corpusLoader

TCCorpusLoader corpusLoader

the loader for the corpus

distributionsDataForTermsOfInterest

TCCorpusDistributionsData distributionsDataForTermsOfInterest

contains the mean and standard deviation of some gaussian distributions

distributionsDataCorpus

TCCorpusDistributionsData distributionsDataCorpus

contains the mean and standard deviation of some gaussian distributions

precisionRecallResults

java.util.ArrayList[] precisionRecallResults

holds all the (recall, precision) values for the precision and recall graph

batchEvaluationResults

java.util.ArrayList batchEvaluationResults

holds all the batch evaluations made so far

parentFrame

javax.swing.JFrame parentFrame

the parent frame (used for progress monitoring etc...)

Constructor Detail

TCExperiment

public TCExperiment()

constructor. set all the values to standard and initialize the classes

Method Detail

setCorpusName

public void setCorpusName(java.lang.String name)

sets the label of the corpus

Parameters:

name - the label of the corpus

setParentJFrame

public void setParentJFrame(javax.swing.JFrame frame)

sets the JFrame this class was created from (used for progress monitoring etc.)

Parameters:

frame - the JFrame object

setKNNk

public void setKNNk(int k)

sets the k value for the kNN classifier

Parameters:

k - number of nearest neighbors considered

setNaiveBayesThreshold

public void setNaiveBayesThreshold(double t)

sets the threshold for the naive bayes classifier

Parameters:

t -

setKNNThreshold

public void setKNNThreshold(double t)

sets the threshold for the naive bayes classifier

Parameters:

t -

setWeightFunctionIndex

public void setWeightFunctionIndex(int wfindex)

sets the weighting function which should be used for the dimensionality reduction step

Parameters:

wfindex - the index of the weighting function

setWeightFunctionSumMethodIndex

public void setWeightFunctionSumMethodIndex(int wfsmi)

sets the method of sumation for the weighting function currently in use

Parameters:

wfsmi - the index of the sumation method

getWeightFunctionIndex

public int getWeightFunctionIndex()

returns the weight function index currently in use

Returns:

the index of the weighting function currently in use

getWeightFunctionSumMethodIndex

public int getWeightFunctionSumMethodIndex()

returns the index of the weight function sumation method

Returns:

index of the sumation method

getNumberOfDocuments

public int getNumberOfDocuments()

returns the number of documents in the experiment corpus

Returns:

number of documents in the loaded corpus

getNumberOfLabeledDocuments

public int getNumberOfLabeledDocuments()

returns the number of documents which were assigned to at least one category

Returns:

the number of labeled documents in the corpus

getNumberOfCategories

public int getNumberOfCategories()

returns the number of categories in the experiment corpus

Returns:

the number of categories in the currently loaded corpus

getNumberOfUniqueTerms

public int getNumberOfUniqueTerms()

returns the number of distinct (unique) terms in the corpus

Returns:

the number of unique terms

getNumberOfProcessedWords

public int getNumberOfProcessedWords()

the number of processed words (the number of words which were sorted into the term vectors by the loadCorpus algorithms

Returns:

the number of processed words

getNumberOfWordsPerDocument

public double getNumberOfWordsPerDocument()

returns the average number of words per document in the corpus

Returns:

average number of words per documents

getNumberOfTestDocuments

public int getNumberOfTestDocuments()

returns the number of test documents currently loaded

Returns:

the number of test documents loaded

getNumberOfTestDocumentsWithLabel

public int getNumberOfTestDocumentsWithLabel()

returns the number of test documents with at least one label (which are essigned to at least one category)

Returns:

the number of test documents loaded

loadStopWordsString

public void loadStopWordsString()

loads the file with all the stop words and saves it in a String for later use

getStopWordsString

public java.lang.String getStopWordsString()

loadCorpus

public void loadCorpus(boolean useSWR,
                       int stemMethod)

loads one of the implemented corpora. To use a different corpus it is necessary to write a load-function for this corpus

Parameters:

useSWR - indicates if the stopword removal should be applied

computeWeightValues

public boolean computeWeightValues(int minimumDFTFThreshold,
                                   javax.swing.ProgressMonitor progressMonitor)

computes the weight values for the experiment corpus

Parameters:

progressMonitor - show progress with a progressMonitor

Returns:

indicates if the algorithm could finish without being interrupted

reduceFeatureSpaceDimension

public double reduceFeatureSpaceDimension(int fSD,
                                          int minimumDFTFFrequency)

reduces the feature space and calculates the distributions and he tfidf values for all the documents in the corpus

Parameters:

fSD - the dimensionality the feature space should have after the reduction step

classifyDocumentAndProvideReport

public java.lang.String classifyDocumentAndProvideReport(java.lang.String path,
                                                         int classifierIndex)

classifies a single document and provides a textual report

Parameters:

path - the path to the document to classify

Returns:

the textual report to pass to the console

trainKNNClassifier

public void trainKNNClassifier()

calculates all the kNN similarity scores for every test documents loaded by the corpusLoader. Takes time!!

trainNaiveBayesClassifier

public void trainNaiveBayesClassifier()

trains the naive bayes classifier (calculates the category probabilities for every document) results are saved in the TCCorpus class for later processing

evaluateClassification

public TCEvaluationResult evaluateClassification(int classifierIndex)

Evaluates the kNN or NBB classificator with the current settings and calculates recall and precision. The evaluation considers the current settings and tests all test document once for one particular setting (using a single threshold given by the user)

Returns:

object which holds the evaluation information

computeClassifierAccuracyKNN

public TCAccuracyResult computeClassifierAccuracyKNN(double stepSize)

computes the accuracy (break even, 11pt precision, precision-recall values, etc.) for the kNN classifier.

Returns:

computeClassifierAccuracyNBB

public TCAccuracyResult computeClassifierAccuracyNBB(double stepSize)

computes the accuracy (break even, precision-recall values) for the naive bayes classifier.

Returns:

evaluateKNNClassificationForThreshold

public TCEvaluationResult evaluateKNNClassificationForThreshold(double threshold,
                                                                boolean display)

Evaluates the kNN classificator with the current settings (and one single threshold) and calculates recall and precision. Will be reduced to one routine for all classifiers.

Returns:

object which holds the evaluation information

evaluateNBBClassificationForThreshold

public TCEvaluationResult evaluateNBBClassificationForThreshold(double threshold,
                                                                boolean display)

Evaluates the naive bayes bernoulli classificator with the current settings (and one single threshold) and calculates recall and precision. Will be reduced to one routine for all classifiers.

Returns:

object which holds the evaluation information

evaluateClassificationBatch

public void evaluateClassificationBatch()

Evaluates (precision and/or recall) one or more classifiers for different types of setting (weighting function, properties of the classificatiob algorithm, number of unique terms, etc.) and fills a TCBatchResult object which can be saved to disk or used for plotting the curves... HAS NOT BEEN FINISHED YET

fillBatchEvaluationDataTable

public TCTableData fillBatchEvaluationDataTable()

creates the table with the batch evaluation data

Returns:

fillCategoryDataTable

public TCTableData fillCategoryDataTable()

resembles the data of every category (number of document, etc.)

Returns:

fillDocumentDataTable

public TCTableData fillDocumentDataTable(int property)

fills the table with properties of all the documents

Returns:

instance of TCTableData for further processing

fillTermDataTable

public TCTableData fillTermDataTable()

shows the properties of all the terms in the terms of interest arraylist (which are the terms after the dimensionality reduction step)

Returns:

createGDGraph

public Graph2D createGDGraph(int distributionIndex)

creates the gaussian distributions graph for later use

Returns:

the graph to display

createPrecisionRecallGraph

public Graph2D createPrecisionRecallGraph(int classifierIndex)

creates the precision<->recall graph based on an ArrayList with evaluation results for many different thresholds

Returns:

createEvaluationGraph

public Graph2D createEvaluationGraph(int evaluationNumber)

creates the batch evaluation graph where different setting (e.g. different weighting functions for different dimensionalities) are plottet

Parameters:

evaluationNumber - indicates the index in the batchevaluation arraylist currently only one batch evaluation can be saved and loaded (index = 0)

Returns:

saveBatchEvaluationResult

public void saveBatchEvaluationResult(java.lang.String path,
                                      int resultIndex)
                               throws java.io.IOException

saves a batch evaluation to the given path

Parameters:

path - the path of the file to save the batch evaluation result in

resultIndex -

Throws:

java.io.IOException

loadBatchEvaluationResult

public void loadBatchEvaluationResult(java.lang.String path)
                               throws java.io.IOException,
                                      java.lang.ClassNotFoundException

Throws:

java.io.IOException

java.lang.ClassNotFoundException

saveClassificatorScoring

public void saveClassificatorScoring(java.lang.String path,
                                     int classifierIndex)
                              throws java.io.IOException

saves the classificator scorings for later use. Saves alot of processing time...

Parameters:

path - the path of the file to save the scoring in

classifierIndex - the index of the classifier the scoring is saved for

Throws:

java.io.IOException

loadClassificatorScoring

public void loadClassificatorScoring(java.lang.String path,
                                     int classifierIndex)
                              throws java.io.IOException,
                                     java.lang.ClassNotFoundException

loads a scoring file.

Parameters:

path - the path of the file the scoring is loaded from

classifierIndex - the index of the classifer for which the scoring is loaded

Throws:

java.io.IOException

java.lang.ClassNotFoundException

exportDocumentVectors

public void exportDocumentVectors(java.lang.String path,
                                  int documentType)
                           throws java.io.IOException

Throws:

java.io.IOException