edu.illinois.cs.cogcomp.lbjava.IR

Class LearningClassifierExpression

java.lang.Object

edu.illinois.cs.cogcomp.lbjava.IR.ASTNode

edu.illinois.cs.cogcomp.lbjava.IR.ClassifierExpression

edu.illinois.cs.cogcomp.lbjava.IR.LearningClassifierExpression

All Implemented Interfaces:

CodeGenerator

public class LearningClassifierExpression
extends ClassifierExpression

This class represents expressions that specify classifiers that learn.

Author:

Nick Rizzolo

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	LearningClassifierExpression.Clause This class represents a clause in a LearningClassifierExpression.

Field Summary

Fields

Modifier and Type	Field and Description
Constant	alpha (¬ø) The desired confidence level for cross validation's confidence interval output; argument to alpha, which can only be specified when cval is also specified.
int	alphaClauses Counts the number of alpha clauses, for error detection.
StringBuffer	attributeString A string representation of the return type information for each feature.
boolean	checkDiscreteValues The SemanticAnalysis pass will let this LearningClassifierExpression know if the features it generates need to be checked for appropriateness in the context of the enclosing ClassifierAssignment by setting this flag.
int	cvalClauses Counts the number of cval clauses for error detection.
static double	defaultAlpha If no alpha clause appears during cross validation, this constant is used.
static InstanceCreationExpression	defaultDiscreteLearner If no learning algorithm is specified to learn a discrete classifier, this learner is used.
static String	defaultPreExtract If no preExtract clause appears in the sources, this constant is used.
static InstanceCreationExpression	defaultRealLearner If no learning algorithm is specified to learn a real classifier, this learner is used.
int	encodingClauses Counts the number of encoding clauses for error detection.
int	evaluateClauses Counts the number of evaluate clauses for error detection.
Expression	evaluation (ø) Tells this learning classifier how to produce a prediction during evaluation; argument to evaluate.
ClassifierExpression	extractor (¬ø) The classifier that does feature extraction for this classifier; argument to using.
Constant	featureEncoding (ø) The encoding that the generated classifier will use when storing string data in features.
Integer	featuresStatus The revision status of the LCE's features node.
int	fromClauses Counts the number of from clauses for error detection.
Constant	K (ø) Represents the integer number of subsets to be used in k-fold cross validation; first argument to cval.
ClassifierExpression	labeler (ø) The classifier this learning classifier gets its labels from.
InstanceCreationExpression	learnerConstructor (ø) Tells this learning classifier how to construct its learning algorithm; argument to with.
Name	learnerName (ø) The name of the learner for this classifier; first argument to with.
Block	learnerParameterBlock (ø) A block of statements that set parameters of the learner for this classifier; second argument to with.
Integer	learningStatus The revision status of the LCE's learning node.
boolean	onlyCodeGeneration This flag is set true iff the changes to the learner's LBJava specification require its Java translation to be regenerated and nothing more.
LinkedList<ParameterSet>	parameterSets A list of the ParameterSet objects that appear in the argument list.
InstanceCreationExpression	parser (ø) Tells this learning classifier how to get its training data; argument to from.
Constant	preExtract (¬ø) A Boolean or string value indicating how feature vectors are to be pre-extracted; argument to preExtract.
int	preExtractClauses Counts the number of preExtract clauses for error detection.
Constant	previousPruneCountType (ø) The contents of pruneCountType on the previous run of the compiler, if any.
Constant	progressOutput (ø) Integer specifying how often (in examples) to give the user a progress update during training; argument to progressOutput.
int	progressOutputClauses Counts the number of progressOutput clauses, for error detection.
Constant	pruneCountType (ø) Whether to use "global" or "perClass" feature pruning.
Integer	pruneStatus The revision status of the LCE's prune node.
Constant	pruneThreshold (ø) The feature pruning threshold.
Constant	pruneThresholdType (ø) Whether to use "count" or "percent" counting for feature pruning.
Expression	rounds (ø) Represents the integer number of training repetitions; augments the from clause.
FoldParser.SplitPolicy	splitPolicy (ø) Dictates how the training data will be split into subsets for use by cross validation; second argument to cval.
int	startingRound Training starts from this round number.
int	testFromClauses Counts the number of testFrom clauses for error detection.
InstanceCreationExpression	testingMetric (ø) Determines how the user wishes cross-validation to test its performance; argument to testingMetric.
int	testingMetricClauses Counts the number of testingMetric clauses, for error detection.
InstanceCreationExpression	testParser (ø) Tells this learning classifier how to get its testing data; argument to testFrom.
int	usingClauses Counts the number of using clauses for error detection.
int	withClauses Counts the number of with clauses for error detection.

Fields inherited from class edu.illinois.cs.cogcomp.lbjava.IR.ClassifierExpression

argument, cacheIn, comment, name, parenthesized, returnType, singleExampleCache

Fields inherited from class edu.illinois.cs.cogcomp.lbjava.IR.ASTNode

byteOffset, line, nodeID, symbolTable

Constructor Summary

Constructors

Constructor and Description
LearningClassifierExpression(ClassifierExpression l, ClassifierExpression ext, InstanceCreationExpression p, Expression r, InstanceCreationExpression a, Name ln, Block pb, Constant enc, InstanceCreationExpression tp, Expression e, Constant k, LinkedList<ParameterSet> prms, FoldParser.SplitPolicy s, InstanceCreationExpression t, Constant al, Constant pre, Constant pro, Constant pct, Constant ptt, Constant pt, Integer ls, Integer fs, Integer ps, StringBuffer at, boolean cdv) Full constructor.
LearningClassifierExpression(ClassifierExpression l, LinkedList<LearningClassifierExpression.Clause> cl, int line, int byteOffset) Parser's supervised learning constructor.
LearningClassifierExpression(LinkedList<LearningClassifierExpression.Clause> cl, int line, int byteOffset) Parser's unsupervised learning constructor.

Method Summary

Modifier and Type	Method and Description
Object	clone() Creates a new object with the same primitive data, and recursively creates new member data objects as well.
boolean	equals(Object o) Indicates whether some other object is "equal to" this one.
int	hashCode() Returns a hash code value for java hash structures.
ASTNodeIterator	iterator() Returns an iterator used to successively access the children of this node.
void	runPass(Pass pass) Ensures that the correct run() method is called for this type of node.
StringBuffer	shallow() Creates a StringBuffer containing a shallow representation of this ASTNode.
void	write(StringBuffer buffer) Writes a string representation of this ASTNode to the specified buffer.

Methods inherited from class edu.illinois.cs.cogcomp.lbjava.IR.ClassifierExpression

getLine, getName, setCacheIn

Methods inherited from class edu.illinois.cs.cogcomp.lbjava.IR.ASTNode

toString

Methods inherited from class java.lang.Object

finalize, getClass, notify, notifyAll, wait, wait, wait

Field Detail

defaultDiscreteLearner

public static final InstanceCreationExpression defaultDiscreteLearner

If no learning algorithm is specified to learn a discrete classifier, this learner is used.

defaultRealLearner

public static final InstanceCreationExpression defaultRealLearner

If no learning algorithm is specified to learn a real classifier, this learner is used.

defaultAlpha

public static final double defaultAlpha

If no alpha clause appears during cross validation, this constant is used.

See Also:

Constant Field Values

defaultPreExtract

public static final String defaultPreExtract

If no preExtract clause appears in the sources, this constant is used.

See Also:

Constant Field Values

featuresStatus

public Integer featuresStatus

The revision status of the LCE's features node.

pruneStatus

public Integer pruneStatus

The revision status of the LCE's prune node.

learningStatus

public Integer learningStatus

The revision status of the LCE's learning node.

onlyCodeGeneration

public boolean onlyCodeGeneration

This flag is set true iff the changes to the learner's LBJava specification require its Java translation to be regenerated and nothing more.

labeler

public ClassifierExpression labeler

(ø) The classifier this learning classifier gets its labels from.

extractor

public ClassifierExpression extractor

(¬ø) The classifier that does feature extraction for this classifier; argument to using.

usingClauses

public int usingClauses

Counts the number of using clauses for error detection.

featureEncoding

public Constant featureEncoding

(ø) The encoding that the generated classifier will use when storing string data in features.

encodingClauses

public int encodingClauses

Counts the number of encoding clauses for error detection.

parser

public InstanceCreationExpression parser

(ø) Tells this learning classifier how to get its training data; argument to from.

fromClauses

public int fromClauses

Counts the number of from clauses for error detection.

rounds

public Expression rounds

(ø) Represents the integer number of training repetitions; augments the from clause.

startingRound

public int startingRound

Training starts from this round number.

pruneCountType

public Constant pruneCountType

(ø) Whether to use "global" or "perClass" feature pruning.

pruneThresholdType

public Constant pruneThresholdType

(ø) Whether to use "count" or "percent" counting for feature pruning.

pruneThreshold

public Constant pruneThreshold

(ø) The feature pruning threshold.

previousPruneCountType

public Constant previousPruneCountType

(ø) The contents of pruneCountType on the previous run of the compiler, if any.

learnerConstructor

public InstanceCreationExpression learnerConstructor

(ø) Tells this learning classifier how to construct its learning algorithm; argument to with. This variable should not contain a non-null value if learnerName and learnerParameterBlock contain non-null values.

learnerName

public Name learnerName

(ø) The name of the learner for this classifier; first argument to with. If this variable contains a non-null value, learnerParameterBlock must also be non-null, and learnerConstructor must be null.

learnerParameterBlock

public Block learnerParameterBlock

(ø) A block of statements that set parameters of the learner for this classifier; second argument to with. If this variable contains a non-null value, learnerName must also be non-null, and learnerConstructor must be null.

withClauses

public int withClauses

Counts the number of with clauses for error detection.

testParser

public InstanceCreationExpression testParser

(ø) Tells this learning classifier how to get its testing data; argument to testFrom.

testFromClauses

public int testFromClauses

Counts the number of testFrom clauses for error detection.

evaluation

public Expression evaluation

(ø) Tells this learning classifier how to produce a prediction during evaluation; argument to evaluate.

evaluateClauses

public int evaluateClauses

Counts the number of evaluate clauses for error detection.

parameterSets

public LinkedList<ParameterSet> parameterSets

A list of the ParameterSet objects that appear in the argument list.

K

public Constant K

(ø) Represents the integer number of subsets to be used in k-fold cross validation; first argument to cval.

splitPolicy

public FoldParser.SplitPolicy splitPolicy

(ø) Dictates how the training data will be split into subsets for use by cross validation; second argument to cval.

cvalClauses

public int cvalClauses

Counts the number of cval clauses for error detection.

testingMetric

public InstanceCreationExpression testingMetric

(ø) Determines how the user wishes cross-validation to test its performance; argument to testingMetric.

testingMetricClauses

public int testingMetricClauses

Counts the number of testingMetric clauses, for error detection.

alpha

public Constant alpha

(¬ø) The desired confidence level for cross validation's confidence interval output; argument to alpha, which can only be specified when cval is also specified.

alphaClauses

public int alphaClauses

Counts the number of alpha clauses, for error detection.

preExtract

public Constant preExtract

(¬ø) A Boolean or string value indicating how feature vectors are to be pre-extracted; argument to preExtract. Possible values are false, true, "false", "true", "none", "memory", "disk", and "diskZip".

preExtractClauses

public int preExtractClauses

Counts the number of preExtract clauses for error detection.

progressOutput

public Constant progressOutput

(ø) Integer specifying how often (in examples) to give the user a progress update during training; argument to progressOutput.

progressOutputClauses

public int progressOutputClauses

Counts the number of progressOutput clauses, for error detection.

checkDiscreteValues

public boolean checkDiscreteValues

The SemanticAnalysis pass will let this LearningClassifierExpression know if the features it generates need to be checked for appropriateness in the context of the enclosing ClassifierAssignment by setting this flag.

attributeString

public StringBuffer attributeString

A string representation of the return type information for each feature. This information is crucial in the construction of WEKA classifiers.

Its format follows this convention: "<type>_<name>[_<value-list>]:<type>_<name>[_<value-list>:[...]]"

<type> can be either "num", "str", or "nom", representing numerical, string, and nominal attributes respectively.

Numerical and string attribute encodings do not need a value-list, while Nominal attribute encodings are required to contain a value list. Examples:

• Just a numerical attribute named Dan: "num_Dan:"
• A numerical attribute named Dan and a string attribute named Nick: "num_Dan:str_Nick:"
• A numerical attribute named Dan, a string attribute named Nick, and a nominal attribute named Arindam which can take the values "Cool", "Uncool", or "Kinda Cool": "num_Dan:str_Nick:nom_Arindam_\"Cool\",\"Uncool\",\"Kinda Cool\":"

Constructor Detail

LearningClassifierExpression

public LearningClassifierExpression(ClassifierExpression l,
                                    ClassifierExpression ext,
                                    InstanceCreationExpression p,
                                    Expression r,
                                    InstanceCreationExpression a,
                                    Name ln,
                                    Block pb,
                                    Constant enc,
                                    InstanceCreationExpression tp,
                                    Expression e,
                                    Constant k,
                                    LinkedList<ParameterSet> prms,
                                    FoldParser.SplitPolicy s,
                                    InstanceCreationExpression t,
                                    Constant al,
                                    Constant pre,
                                    Constant pro,
                                    Constant pct,
                                    Constant ptt,
                                    Constant pt,
                                    Integer ls,
                                    Integer fs,
                                    Integer ps,
                                    StringBuffer at,
                                    boolean cdv)

Full constructor. The line and byte offset are set to -1.

Parameters:

l - The expression representing the labeler classifier.

ext - Representation of the extractor classifier.

p - The expression representing the parser applied to data.

r - The constant representing the number of training repetitions.

a - A learning algorithm instance creation expression.

ln - The name of a learning algorithm.

pb - A block of statements used to set learning algorithm parameters.

enc - The expression representing the feature encoding used in this learner's lexicon.

tp - The expression representing the test parser.

e - The expression used to compute a prediction.

k - The number of folds for cross validation.

prms - The sets of parameter values used in tuning.

s - The method used to split the data for cross validation.

t - Determines how the user wishes cross-validation to test its performance.

al - The cross validation confidence interval width.

pre - A Boolean indicating whether example vectors will be pre-extracted.

pro - An integer indicating how often progress updates will be output.

pct - The prune type ("global" or "perClass").

ptt - The prune count type ("count" or "percent").

pt - The prune count threshold.

ls - The "learning status" node, set by RevisionAnalysis.

fs - The "features status" node, set by RevisionAnalysis.

ps - The "prune status" node, set by RevisionAnalysis.

at - The WEKA attribute string.

cdv - Value for checkDiscreteValues.

LearningClassifierExpression

public LearningClassifierExpression(LinkedList<LearningClassifierExpression.Clause> cl,
                                    int line,
                                    int byteOffset)

Parser's unsupervised learning constructor.

Parameters:

cl - A list of clauses.

line - The line on which the source code represented by this node is found.

byteOffset - The byte offset from the beginning of the source file at which the source code represented by this node is found.

LearningClassifierExpression

public LearningClassifierExpression(ClassifierExpression l,
                                    LinkedList<LearningClassifierExpression.Clause> cl,
                                    int line,
                                    int byteOffset)

Parser's supervised learning constructor.

Parameters:

l - The expression representing the labeler classifier.

cl - A list of clauses.

line - The line on which the source code represented by this node is found.

byteOffset - The byte offset from the beginning of the source file at which the source code represented by this node is found.

Method Detail

hashCode

public int hashCode()

Returns a hash code value for java hash structures.

Overrides:

hashCode in class Object

Returns:

A hash code for this object.

equals

public boolean equals(Object o)

Indicates whether some other object is "equal to" this one.

Overrides:

equals in class Object

Returns:

true iff this object is the same as the argument.

iterator

public ASTNodeIterator iterator()

Returns an iterator used to successively access the children of this node.

Specified by:

iterator in class ASTNode

Returns:

An iterator used to successively access the children of this node.

clone

public Object clone()

Creates a new object with the same primitive data, and recursively creates new member data objects as well.

Overrides:

clone in class ASTNode

Returns:

The clone node.

runPass

public void runPass(Pass pass)

Ensures that the correct run() method is called for this type of node.

Specified by:

runPass in class ASTNode

Parameters:

pass - The pass whose run() method should be called.

write

public void write(StringBuffer buffer)

Writes a string representation of this ASTNode to the specified buffer. The representation written is parsable by the LBJava compiler, but not very readable.

Specified by:

write in class ASTNode

Parameters:

buffer - The buffer to write to.

shallow

public StringBuffer shallow()

Creates a StringBuffer containing a shallow representation of this ASTNode.

Specified by:

shallow in interface CodeGenerator

Specified by:

shallow in class ClassifierExpression

Returns:

A StringBuffer containing a shallow text representation of the given node.