edu.harvard.seas.iis.abilities.classify
Class MovementClassifier

java.lang.Object
  edu.harvard.seas.iis.abilities.classify.MovementClassifier

public class MovementClassifier
extends java.lang.Object

A front end class with high level tools for classifying pointing movements as deliberate or distracted; uses pre-trained classifiers to perform the classification so it is all ready to go.

Author:

kgajos

Field Summary

static java.lang.String	DELIBERATE_PROBABILITY_KEY
static java.lang.String	PREDICTED_CLASS_KEY
static java.lang.String	REPORT_HEADER

Constructor Summary

MovementClassifier()
Initializes the class and loads the default serialized versions of the classifiers from the following location in the classpath: edu/harvard/seas/iis/abilities/classify/resources/

MovementClassifier(java.lang.String classifierDirectory)
Initializes the class and loads serialized classifiers from the location indicated in classifierDirectory variable

Method Summary

void	classifyMovements(java.util.Collection<Movement> movements, java.util.Vector<Movement> deliberate, java.util.Vector<Movement> distracted, boolean useTargetInformation) Takes movements and classifies them; classified movements are added to the deliberate and distracted vectors
DataSet	classifyMovements(DataSet movements, boolean useTargetInformation) Assumes the movements data set has "Predicted class" and "Prediction probability" attributes present.
double	getDeliberateProbability(weka.core.Instance movement, boolean useTargetInformation) Returns the probability that the movement was deliberate
double	getDeliberateProbability(Movement movement, boolean useTargetInformation) Returns the probability that the movement was deliberate; also modifies the movement object by adding two pieces of meta data: "deliberate probability" and "predicted class"
static void	main(java.lang.String[] args)
java.util.Vector<Movement>	parseAndClassify(java.io.File[] files, MovementFilter filter, Parser parser, boolean useTargetInformation, boolean returnParsedMovements, java.util.Vector<java.lang.String> report) Reads log files, uses the parser to extract movements out of the log files, uses the filter to select only the subset of the movements that are of interest, and classifies them.

Methods inherited from class java.lang.Object

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

Field Detail

DELIBERATE_PROBABILITY_KEY

public static final java.lang.String DELIBERATE_PROBABILITY_KEY

See Also:

Constant Field Values

PREDICTED_CLASS_KEY

public static final java.lang.String PREDICTED_CLASS_KEY

See Also:

Constant Field Values

REPORT_HEADER

public static final java.lang.String REPORT_HEADER

See Also:

Constant Field Values

Constructor Detail

MovementClassifier

public MovementClassifier()

Initializes the class and loads the default serialized versions of the classifiers from the following location in the classpath: edu/harvard/seas/iis/abilities/classify/resources/

MovementClassifier

public MovementClassifier(java.lang.String classifierDirectory)

Initializes the class and loads serialized classifiers from the location indicated in classifierDirectory variable

Parameters:

classifierDirectory - the location where the serialized classifiers may be found

Method Detail

classifyMovements

public void classifyMovements(java.util.Collection<Movement> movements,
                              java.util.Vector<Movement> deliberate,
                              java.util.Vector<Movement> distracted,
                              boolean useTargetInformation)
                       throws java.lang.Exception

Takes movements and classifies them; classified movements are added to the deliberate and distracted vectors

Parameters:

movements - movement objects to be classified

deliberate - movements classified as deliberate will be added to this vector

distracted - movements classified as distracted will be added to this vector

useTargetInformation - if set to true, the classifier will make use of features that rely on the knowledge of the target size; if set to false, the classification will be done in a target agnostic manner

Throws:

java.lang.Exception

getDeliberateProbability

public double getDeliberateProbability(Movement movement,
                                       boolean useTargetInformation)
                                throws java.lang.Exception

Returns the probability that the movement was deliberate; also modifies the movement object by adding two pieces of meta data: "deliberate probability" and "predicted class"

Parameters:

movement - Movement to classify

useTargetInformation - if set to true, the classifier will make use of features that rely on the knowledge of the target size; if set to false, the classification will be done in a target agnostic manner

Returns:

probability that the movement was deliberate or -1 if the movement could not be classified

Throws:

java.lang.Exception

getDeliberateProbability

public double getDeliberateProbability(weka.core.Instance movement,
                                       boolean useTargetInformation)
                                throws java.lang.Exception

Returns the probability that the movement was deliberate

Parameters:

movement - movement already encoded as a Weka Instance, transformed to include additional features, and normalized!

useTargetInformation - if set to true, the classifier will make use of features that rely on the knowledge of the target size; if set to false, the classification will be done in a target agnostic manner

Returns:

Throws:

java.lang.Exception

classifyMovements

public DataSet classifyMovements(DataSet movements,
                                 boolean useTargetInformation)
                          throws java.lang.Exception

Assumes the movements data set has "Predicted class" and "Prediction probability" attributes present. This method classifies each movement and sets the value of these two attributes and returns such annotated data set

Parameters:

movements - movements encoded as Weka Intances; they have to have had additional features already computed and they have to have been normalized before being passed here

useTargetInformation -

Returns:

Throws:

java.lang.Exception

parseAndClassify

public java.util.Vector<Movement> parseAndClassify(java.io.File[] files,
                                                   MovementFilter filter,
                                                   Parser parser,
                                                   boolean useTargetInformation,
                                                   boolean returnParsedMovements,
                                                   java.util.Vector<java.lang.String> report)
                                            throws java.lang.Exception

Reads log files, uses the parser to extract movements out of the log files, uses the filter to select only the subset of the movements that are of interest, and classifies them. For now it just prints an aggregate report to the stdout

Parameters:

files - log files from which the movements are to be extracted

filter - filter for selecting a subset of the movements to be classified; if null, all movements get classified

parser - instance of Parser for extracting Movements from the log

useTargetInformation - if set to true, the classifier will make use of features that rely on the knowledge of the target size; if set to false, the classification will be done in a target agnostic manner

returnParsedMovements - will return a vector of Movement objects if this parameter is set to true; will return an empty vector otherwise; useful, if all that is needed is the report and not the movements themselves

report - a string report will be appended to this variable

Throws:

java.lang.Exception

main

public static void main(java.lang.String[] args)
                 throws java.lang.Exception

Parameters:

args -

Throws:

java.lang.Exception