changelog-3-2

这是一个由java实现的数据挖掘

	classifiers/SMO.java (1.19), classifiers/ThresholdSelector.java
	(1.13), classifiers/VotedPerceptron.java (1.7),
	classifiers/m5/M5Prime.java (1.11), clusterers/SimpleKMeans.java
	(1.2): Updated users of Filters so that incremental use is followed
	by a call to batchFinished() -- this gives memory optimization in
	the case of datasets with string attributes as references to old
	strings are flushed.

2000-12-10 13:58  trigg

	* weka/filters/: AddFilter.java (1.10), AllFilter.java (1.5),
	AttributeExpressionFilter.java (1.5), AttributeFilter.java (1.10),
	AttributeSelectionFilter.java (1.16), AttributeTypeFilter.java
	(1.4), CopyAttributesFilter.java (1.5), DiscretizeFilter.java
	(1.10), EmptyAttributeFilter.java (1.3), Filter.java (1.12),
	FirstOrderFilter.java (1.8), InstanceFilter.java (1.5),
	MakeIndicatorFilter.java (1.8), MergeTwoValuesFilter.java (1.6),
	NominalToBinaryFilter.java (1.12), NonSparseToSparseFilter.java
	(1.3), NormalizationFilter.java (1.6), NullFilter.java (1.4),
	NumericToBinaryFilter.java (1.3), NumericTransformFilter.java
	(1.8), ObfuscateFilter.java (1.4), RandomizeFilter.java (1.4),
	ReplaceMissingValuesFilter.java (1.8), ResampleFilter.java (1.6),
	SparseToNonSparseFilter.java (1.3), SplitDatasetFilter.java (1.7),
	SpreadSubsampleFilter.java (1.5), StringToNominalFilter.java (1.3),
	StringToWordVectorFilter.java (1.4), SwapAttributeValuesFilter.java
	(1.7), TimeSeriesDeltaFilter.java (1.4),
	TimeSeriesTranslateFilter.java (1.5): Huge cleanup here. Filters
	now work safely with string attributes. Previously, Bad Things
	would happen and the string values would get lost or point to the
	wrong strings when passing through filters.
	
	There is one case (not applicable all known filters) yet to handle,
	see comments in Filter.inputFormat.

2000-12-10 13:53  trigg

	* weka/core/Instances.java (1.25): Added a method for quickly
	removing all instances from the dataset.

2000-12-10 13:52  trigg

	* weka/core/Instance.java (1.9): copy() now points the copied
	instance at the same dataset as the original instance. This makes
	more sense. If you want the old behaviour, use the equivalent
	constructor.

2000-12-10 13:51  trigg

	* weka/core/Attribute.java (1.14): Adding string values no longer
	checks for duplicates (since it's assumed string values are
	typically unique, otherwise they'd be nominal).
	
	Added a method for efficiently copying string values from one
	attribute to another.

2000-12-09 14:13  trigg

	* weka/: classifiers/FilteredClassifier.java (1.8),
	core/Instances.java (1.24), filters/StringToWordVectorFilter.java
	(1.3): Moved common code into Instances.

2000-12-09 14:13  trigg

	* weka/classifiers/MultiClassClassifier.java (1.13): Less
	restrictive of the datasets it accepts.

2000-12-09 11:42  trigg

	* weka/: associations/Associator.java (1.3),
	attributeSelection/ASEvaluation.java (1.7),
	classifiers/Classifier.java (1.7), clusterers/Clusterer.java (1.7):
	Changed makeCopies methods to use Serialization when copying. Now
	copies will also contain any model built by the input object.
	However, this method now works when making copies of objects that
	were not previously able to be created by the command line.

2000-12-08 16:13  trigg

	* weka/classifiers/evaluation/ThresholdCurve.java (1.10): Added
	routine for calculating area under ROC curve.

2000-12-08 15:30  trigg

	* weka/filters/MakeIndicatorFilter.java (1.7): Changed how output
	indicator values are named.

2000-12-08 15:25  trigg

	* weka/classifiers/ThresholdSelector.java (1.12): Little more
	accepting in the auto-name case. Now just wants class values
	starting with "yes" or "pos" (case-insensitive) or equalling "1".

2000-12-08 15:17  trigg

	* weka/attributeSelection/StartSetHandler.java (1.4): Fixed buggy
	Revision tag.

2000-12-08 15:04  trigg

	* weka/classifiers/ThresholdSelector.java (1.11): Another method
	for picking which class to optimize: value with a positive sounding
	name (yes, pos, positive, 1);

2000-12-08 14:12  trigg

	* weka/filters/SpreadSubsampleFilter.java (1.4): Added an option to
	adjust weights of output instances so that the total weight per
	class can be maintained.

2000-12-08 11:42  trigg

	* weka/classifiers/evaluation/Prediction.java (1.3): Incorrect
	access on the predicted method.

2000-12-08 11:37  trigg

	* weka/classifiers/evaluation/TwoClassStats.java (1.4): Added a
	method to get an equivalent confusion matrix. Perhaps TwoClassStats
	should be changed to use a ConfusionMatrix internally. Advantage:
	loading/saving/applying costs. Disadvantage: slightly slower,
	slightly more memory.

2000-12-08 11:35  trigg

	* weka/classifiers/Evaluation.java (1.38): Added a couple of
	comments.

2000-12-08 11:34  trigg

	* weka/classifiers/CostSensitiveClassifier.java (1.6):
	setCostMatrix and setOnDemandDirectory now set the cost matrix
	source appropriately.

2000-12-08 11:33  trigg

	* weka/classifiers/CostMatrix.java (1.7): A couple of extra
	comments. No actual code change.

2000-12-06 09:46  trigg

	* weka/classifiers/MetaCost.java (1.5): setCostMatrix and
	setOnDemandDirectory now make the appropriate setting to
	setCostMatrixSource. We should probably go through properties in
	other schemes and get the property dependencies updating
	appropriately.

2000-12-05 13:58  trigg

	* weka/classifiers/Evaluation.java (1.37): Partial refactoring of
	the main evaluation routine.

2000-12-04 15:22  trigg

	* weka/classifiers/AttributeSelectedClassifier.java (1.5): Fixed
	small bug in Revision tag.

2000-12-04 15:18  trigg

	* weka/attributeSelection/SymmetricalUncertAttributeEval.java
	(1.9): Fixed small bug in Revision tag.

2000-12-04 15:14  trigg

	* weka/attributeSelection/: ErrorBasedMeritEvaluator.java (1.3),
	RankedOutputSearch.java (1.6): Fixed small bug in Revision tag.

2000-12-04 12:26  trigg

	* weka/classifiers/ThresholdSelector.java (1.10): Added different
	options for choosing which class value to optimize for.

2000-12-04 10:52  trigg

	* weka/classifiers/ThresholdSelector.java (1.9): Converted into a
	DistributionClassifier, where the probability distribution is
	rescaled so that the selected threshold corresponds to 0.5 in the
	output distribution. Essentially means this metaclassifier just
	does two linear transformations on the output of the subclassifier.

2000-12-03 16:32  trigg

	* weka/classifiers/evaluation/ThresholdCurve.java (1.9): Now
	possible to get the names of the attributes used for the various
	measures as static fields.

2000-12-03 16:31  trigg

	* weka/classifiers/evaluation/EvaluationUtils.java (1.4): Can now
	generate NumericPredictions.

2000-12-03 16:30  trigg

	* weka/classifiers/DistributionMetaClassifier.java (1.3): Added a
	useful constructor.

2000-12-03 16:29  trigg

	* weka/classifiers/ThresholdSelector.java (1.8): Removed 120 lines
	of code :-)

2000-11-30 15:48  trigg

	* weka/: Makefile (1.29), core/Makefile (1.9),
	core/converters/AbstractLoader.java (1.1),
	core/converters/ArffLoader.java (1.1),
	core/converters/C45Loader.java (1.1),
	core/converters/CSVLoader.java (1.1),
	core/converters/ConverterUtils.java (1.1),
	core/converters/Loader.java (1.1), core/converters/Makefile (1.1),
	core/converters/SerializedInstancesLoader.java (1.1),
	gui/GenericObjectEditor.props (1.28),
	gui/explorer/PreprocessPanel.java (1.21): Moved weka.converters to
	weka.core.converters Did some renaming of classes Added ability for
	some Loaders to read from an InputStream Implemented
	SerializedInstancesLoader

2000-11-29 16:13  trigg

	* weka/classifiers/FilteredClassifier.java (1.7): The memory
	optimization to remove references to old strings after filtering
	wasn't passing the class attribute through. Fixed.

2000-11-24 13:11  mhall

	* weka/gui/explorer/Explorer.java (1.18): Oops. Changed addPlot to
	setMasterPlot when setting up Visualize panel.

2000-11-24 12:55  mhall

	* weka/gui/explorer/Explorer.java (1.17): Visualize panel is now
	initialized using a PlotData2D object rather than calling
	setInstances.

2000-11-24 12:54  mhall

	* weka/gui/visualize/VisualizePanel.java (1.17): The setInstances
	method does not add an instance number attribute (for the sake of
	the UserClassifier).

2000-11-24 11:40  mhall

	* weka/gui/explorer/: ClassifierPanel.java (1.41),
	AttributeSelectionPanel.java (1.22), ClustererPanel.java (1.28):
	Visualizable results now have instance number attribute.

2000-11-24 11:39  mhall

	* weka/gui/visualize/: PlotData2D.java (1.9), VisualizePanel.java
	(1.16): Made adding an instance number attribute to visualizeable
	instances optional.

2000-11-23 14:16  mhall

	* weka/gui/visualize/PlotData2D.java (1.8): Now preserves original
	relation name (used to get overwritten by the filter that is used
	to add the instance number attribute).

2000-11-22 17:27  trigg

	* weka/classifiers/MultiClassClassifier.java (1.12): Optimization
	when using on two-class problems -- doesn't build a redundant
	model.

2000-11-22 08:54  mhall

	* weka/gui/GenericObjectEditor.props (1.27): Added SimpleKMeans

2000-11-22 08:54  mhall

	* weka/clusterers/: Makefile (1.3), SimpleKMeans.java (1.1):
	Initial import SimpleKMeans. Added to Makefile.

2000-11-20 15:06  mhall

	* weka/gui/visualize/PlotData2D.java (1.7): Now adds an
	"Instance_number" attribute

2000-11-10 16:18  mhall

	* weka/experiment/ClassifierSplitEvaluator.java (1.11): Added kappa
	statistic to the results produced.

2000-11-03 18:11  eibe

	* weka/filters/SplitDatasetFilter.java (1.6): Option to generate
	data that is not stratified even if class index is set.

2000-11-01 16:25  mhall

	* weka/classifiers/AdditiveRegression.java (1.4): Added option for
	setting the maximum number of iterations to perform.

2000-11-01 12:24  eibe

	* weka/classifiers/Evaluation.java (1.36): Included kappa statistic
	in output.

2000-11-01 11:41  mhall

	* weka/classifiers/Prism.java (1.11): Fixed bug that Eibe
	introduced in July last year :-)

2000-10-30 15:49  mhall

	* weka/gui/GenericObjectEditor.props (1.26): Added converter
	C45ToArff.

2000-10-30 10:31  trigg

	* weka/classifiers/MultiClassClassifier.java (1.10): Better output
	-- includes class names rather than just indices.

2000-10-24 12:36  mhall

	* weka/gui/explorer/: ClassifierPanel.java (1.39),
	ClustererPanel.java (1.26), AttributeSelectionPanel.java (1.20):
	Now sets the log object for VisualizePanels that are popped up.

2000-10-24 12:35  mhall

	* weka/gui/visualize/VisualizePanel.java (1.14): Added a test for
	non null log object before trying to write a message to it :-)

2000-10-24 10:38  trigg

	* weka/core/: BinarySparseInstance.java (1.1), Instance.java (1.7),
	SparseInstance.java (1.3): Added first version of an Instance
	optimized for binary data (i.e. less memory requirements). Needs
	more testing. Also required some changes to finality of methods in
	parent classes.

2000-10-20 16:33  mhall

	* weka/gui/experiment/RunPanel.java (1.14): No longer "says waiting
	for remote tasks to finish" if experiment is not a
	RemoteExperiment.

2000-10-20 15:42  mhall

	* weka/classifiers/m5/Node.java (1.3): Fixed bug that prevented the
	last attribute from being considered when making a split at a node.