changelog-3-2

这是一个由java实现的数据挖掘

	add predictions to the vector if the actual class value is unknown.

2001-01-22 16:47  trigg

	* weka/classifiers/ThresholdSelector.java (1.19): buildClassifier
	resets thresholds and bounds between subsequent calls.	Handles
	case where only one class value is present.  No longer removes
	instances with missing class value (it's up to the base learner to
	decide).

2001-01-22 16:46  trigg

	* weka/classifiers/MultiClassClassifier.java (1.20): Fixed bug in
	toString.

2001-01-19 14:19  trigg

	* weka/: associations/Apriori.java (1.8),
	attributeSelection/BestFirst.java (1.18),
	attributeSelection/RaceSearch.java (1.6),
	classifiers/AdditiveRegression.java (1.5),
	classifiers/AttributeSelectedClassifier.java (1.6),
	classifiers/CostSensitiveClassifier.java (1.8),
	classifiers/DecisionTable.java (1.19), classifiers/IBk.java (1.16),
	classifiers/LinearRegression.java (1.10), classifiers/MetaCost.java
	(1.6), classifiers/MultiClassClassifier.java (1.19),
	classifiers/ThresholdSelector.java (1.18), classifiers/j48/J48.java
	(1.19), classifiers/j48/PART.java (1.13),
	classifiers/kstar/KStar.java (1.5), classifiers/m5/M5Prime.java
	(1.13), core/AdditionalMeasureProducer.java (1.4),
	core/Attribute.java (1.16), core/AttributeStats.java (1.4),
	core/BinarySparseInstance.java (1.3), core/Instance.java (1.11),
	core/Instances.java (1.28), core/Matrix.java (1.10),
	core/Range.java (1.8), core/SelectedTag.java (1.4),
	core/SparseInstance.java (1.7), core/Tag.java (1.3),
	core/Utils.java (1.25), experiment/AveragingResultProducer.java
	(1.12), experiment/ClassifierSplitEvaluator.java (1.12),
	experiment/CrossValidationResultProducer.java (1.10),
	experiment/DatabaseResultProducer.java (1.12),
	experiment/LearningRateResultProducer.java (1.3),
	experiment/RandomSplitResultProducer.java (1.12),
	experiment/RegressionSplitEvaluator.java (1.11),
	filters/AttributeTypeFilter.java (1.6): Changed a bunch of
	Exceptions to IllegalArgumentExceptions, and removed newly
	redundant try/catch blocks.

2001-01-18 11:34  trigg

	* weka/core/Instances.java (1.27): Added a check for 0 folds to
	stratify method.

2001-01-18 11:16  trigg

	* weka/classifiers/ThresholdSelector.java (1.17): Just passes data
	through if only one class value found.

2001-01-17 12:08  trigg

	* weka/filters/: AbstractTimeSeriesFilter.java (1.1),
	AddFilter.java (1.13), AttributeFilter.java (1.13), Makefile
	(1.17), StringToWordVectorFilter.java (1.5),
	SwapAttributeValuesFilter.java (1.8), TimeSeriesDeltaFilter.java
	(1.5), TimeSeriesTranslateFilter.java (1.7): Fixed string handling.
	 Fixed a couple of minor bugs found during unit testing.

2001-01-17 12:07  trigg

	* weka/core/SparseInstance.java (1.6): Added some diagnostic
	messages.

2001-01-15 15:54  trigg

	* weka/filters/: InstanceFilter.java (1.7),
	MakeIndicatorFilter.java (1.11), ResampleFilter.java (1.7),
	SparseToNonSparseFilter.java (1.4), SplitDatasetFilter.java (1.8):
	More fixes, mainly related to string value handling.

2001-01-15 15:53  trigg

	* weka/filters/Filter.java (1.19): getOutputFormat() now makes a
	shallow copy of the instances. Shouldn't be too much extra
	overhead, but cache the result if you're concerned.

2001-01-15 15:52  trigg

	* weka/: TODO.dtd (1.1), TODO.xml (1.1): First version of files to
	track things to do. We should be able to process this into other
	formats like html, plaintext etc using XSLT.

2001-01-15 11:34  trigg

	* weka/core/AttributeStats.java (1.3): Added a toString method.

2001-01-15 11:33  trigg

	* weka/core/Range.java (1.7): Changed a bunch of exceptions to be
	unchecked exceptions (since they are thrown as a result of caller
	errors rather than environmental errors).

2001-01-12 16:21  trigg

	* weka/filters/: NominalToBinaryFilter.java (1.13),
	NonSparseToSparseFilter.java (1.4), NormalizationFilter.java (1.7),
	NumericToBinaryFilter.java (1.4), NumericTransformFilter.java
	(1.10), ReplaceMissingValuesFilter.java (1.9): Fixed passing
	through of string attribute values correctly.

2001-01-12 16:04  trigg

	* weka/core/SparseInstance.java (1.5): Made copy() also copy the
	reference to the instance's dataset, like in Instance.copy()

2001-01-09 15:21  trigg

	* weka/filters/Filter.java (1.18): Took out buggy expression in
	string optimization code.

2001-01-09 14:36  eibe

	* weka/core/Attribute.java (1.15): Added hashtable for string
	attributes. New values are only added if not present in hashtable.

2001-01-09 08:20  trigg

	* weka/filters/: AddFilter.java (1.12),
	AttributeExpressionFilter.java (1.7), AttributeFilter.java (1.12),
	AttributeSelectionFilter.java (1.17), AttributeTypeFilter.java
	(1.5), CopyAttributesFilter.java (1.7), DiscretizeFilter.java
	(1.12), EmptyAttributeFilter.java (1.4), Filter.java (1.17),
	FirstOrderFilter.java (1.10), InstanceFilter.java (1.6): Updated to
	work safely with string attributes, and any bugs found during
	writing of JUnit tests fixed.

2001-01-09 08:17  trigg

	* weka/gui/GenericObjectEditor.java (1.23): Fixed the problem where
	you have to load the object twice to get sub-properties to load in,
	although the fix could be better.

2001-01-09 08:15  trigg

	* weka/core/Instance.java (1.10): Added some diagnostic messages.

2001-01-09 08:14  trigg

	* weka/core/Queue.java (1.5): Changed some exceptions to be more in
	line with similar java library classes.

2000-12-29 14:36  trigg

	* weka/gui/GenericObjectEditor.props (1.30): Added
	LearningRateResultProducer

2000-12-22 15:51  trigg

	* weka/classifiers/ThresholdSelector.java (1.16): A few more
	javadocs added. JDE's "check documenation" feature *rocks*.

2000-12-22 15:42  trigg

	* weka/classifiers/ThresholdSelector.java (1.15): Added ability to
	expand the range of probabilities returned by the base learner to
	[0,1] based on values seen during the optimization.
	
	Added documentation available via the gui.

2000-12-20 11:45  trigg

	* weka/classifiers/MultiClassClassifier.java (1.18): Reinserted
	indicator filtering of test instances. Even though classifiers
	shouldn't really be looking at the class attribute of test
	instances, some (such as j48) currently do look at the test
	instance class attribute to find the number of class values.
	
	Maybe we should probably have a policy that classifiers do not
	access the class attribute of test instances (and write test code
	to check this).

2000-12-20 11:26  trigg

	* weka/filters/MakeIndicatorFilter.java (1.10): Added property
	documentation for gui.

2000-12-20 11:23  trigg

	* weka/gui/PropertySheetPanel.java (1.8): Outputs help entries for
	properties that don't contain a .

2000-12-20 11:21  trigg

	* weka/filters/AttributeExpressionFilter.java (1.6): Capitalization
	of tip texts.

2000-12-20 09:49  trigg

	* weka/filters/: AttributeFilter.java (1.11),
	CopyAttributesFilter.java (1.6), DiscretizeFilter.java (1.11),
	FirstOrderFilter.java (1.9), NumericTransformFilter.java (1.9),
	TimeSeriesTranslateFilter.java (1.6): Refactored some common code
	out into Range.java

2000-12-20 09:49  trigg

	* weka/gui/GenericObjectEditor.props (1.29): Added a couple of
	classifiers. Fixed a bug in the entry for DistributionClassifiers
	(trailing whitespace does matter).

2000-12-20 09:48  trigg

	* weka/classifiers/CostSensitiveClassifier.java (1.7): Added
	documentation available via the gui.

2000-12-20 09:47  trigg

	* weka/filters/MakeIndicatorFilter.java (1.9): MakeIndicatorFilter
	can now indicate on any combination of attribute values.

2000-12-20 09:46  trigg

	* weka/core/Range.java (1.6): Added a constructor to initialize the
	Range with a range list.

2000-12-20 09:45  trigg

	* weka/classifiers/MultiClassClassifier.java (1.17): Now does
	error-correcting output codes. Two code generation methods are
	implemented: random codes (with simple tests for code goodness),
	and exhaustive code generation. Also added a bunch of methods for
	providing docs via the gui.

2000-12-19 15:45  trigg

	* weka/gui/PropertySheetPanel.java (1.7): Better help width and
	proper initial caret position.

2000-12-13 18:48  trigg

	* weka/filters/Filter.java (1.16): Removed timing instrumentation
	code. ooops.

2000-12-13 18:28  trigg

	* weka/classifiers/ThresholdSelector.java (1.14): Reduces the
	number of x-val folds if there are not enough training instances.

2000-12-13 18:15  trigg

	* weka/filters/Filter.java (1.15): Fixed that case I mentioned
	before. Input buffering is now string-safe.

2000-12-13 18:09  trigg

	* weka/classifiers/SMO.java (1.21): Changed internal
	m_dontNormalize variable to m_Normalize. Less stressful on the
	brain when looking at
	
	if (m_Normalize)
	
	than
	
	if (!m_dontNormalize)
	
	No change to command-line or gui use.

2000-12-12 16:05  trigg

	* weka/filters/Filter.java (1.14): Smarter about whether and how
	often String references are flushed. Much speedup for incremental
	filtering.

2000-12-12 16:04  trigg

	* weka/clusterers/DistributionMetaClusterer.java (1.3): Took bodgy
	pseudo-WeightedInstancesHandling out of this meta clusterer.  See
	log comments for DistributionMetaClassifier for why.

2000-12-12 16:03  trigg

	* weka/classifiers/: DistributionMetaClassifier.java (1.4),
	FilteredClassifier.java (1.11), MultiClassClassifier.java (1.16),
	RegressionByDiscretization.java (1.11): Took bodgy
	pseudo-WeightedInstancesHandler out of these meta classifiers.	The
	right way to do this is to write a WeightedInstancesMetaClassifier
	and give the user the choice between resampling with/without
	replacement and with varying sizes of post-resampling data.

2000-12-11 10:36  trigg

	* weka/filters/Filter.java (1.13): Performs a memory optimization
	after the last instance is retrieved from the output queue between
	batches.
	
	The recent changes to Filter will require updates to any filters
	you have implemented yourself. The changes are:
	
	In your implementation of inputFormat(Instances) you should call
	super.inputFormat(Instances) as the first line.
	
	If your filter requires buffering instances in the first batch, use
	bufferInput(Instance) to do this. Your implementation of
	batchFinished should also call flushInput() (after processing
	buffered instances if necessary).
	
	Other references to m_InputFormat should be replaced by
	getInputFormat().

2000-12-11 10:23  trigg

	* weka/classifiers/evaluation/ThresholdCurve.java (1.11): Now
	handles non-integer prediction weights properly.

2000-12-11 10:23  trigg

	* weka/filters/AddFilter.java (1.11): Fixed bug introduced by
	Instance.copy() maintaining the dataset reference.

2000-12-10 16:18  trigg

	* weka/core/Instances.java (1.26): Made stringFreeStructure more
	efficient.

2000-12-10 16:16  trigg

	* weka/: attributeSelection/AttributeSelection.java (1.24),
	attributeSelection/PrincipalComponents.java (1.14),
	classifiers/ClassificationViaRegression.java (1.7),
	classifiers/DecisionTable.java (1.18),
	classifiers/FilteredClassifier.java (1.10),
	classifiers/LinearRegression.java (1.9), classifiers/Logistic.java
	(1.9), classifiers/MultiClassClassifier.java (1.15),
	classifiers/RegressionByDiscretization.java (1.10),
	classifiers/SMO.java (1.20), classifiers/VotedPerceptron.java
	(1.8), classifiers/m5/M5Prime.java (1.12),
	clusterers/SimpleKMeans.java (1.3): Reordered batchFinished calls.

2000-12-10 14:02  trigg

	* weka/: attributeSelection/AttributeSelection.java (1.23),
	attributeSelection/PrincipalComponents.java (1.13),
	classifiers/ClassificationViaRegression.java (1.6),
	classifiers/DecisionTable.java (1.17),
	classifiers/FilteredClassifier.java (1.9),
	classifiers/LinearRegression.java (1.8), classifiers/Logistic.java
	(1.8), classifiers/MultiClassClassifier.java (1.14),
	classifiers/RegressionByDiscretization.java (1.9),