Weka-3-5-7 Changelog

Designed for use with Apache Ant.


2007-12-17 01:29 mhall
Can now make use of attribute weights
2007-12-16 22:17 fracpete
moved the remaining dynamic discovery of classes (via ClassDiscovery.find) into GenericPropertiesCreator.props/GenericObjectEditor.props. This should enable environments without the possibility of dynamic class discovery, e.g., Eclipse, to use these classes as well (as long GenericObjectEditor.props is maintained)
2007-12-16 22:17 fracpete
moved the remaining dynamic discovery of classes (via ClassDiscovery.find) into GenericPropertiesCreator.props/GenericObjectEditor.props. This should enable environments without the possibility of dynamic class discovery, e.g., Eclipse, to use these classes as well (as long GenericObjectEditor.props is maintained)
2007-12-13 21:32 mhall
Changed so that only the weka.jar goes into the OS X application as a resource, all other files go into a separate directory in the disk image
2007-12-13 02:45 mhall
branches:  1.1.2;  1.1.4;
Initial import of Mac OS X icon
2007-12-12 21:40 fracpete
branches:  1.1.2;
filter for selection a random subset of attributes, either percentage or absolute number of attributes
2007-12-12 21:03 mhall
Fixed return value for accessor method for sample size.
2007-12-10 20:22 fracpete
max number of rows to display gets only output in the infopanel, if we really hit the max number of rows
2007-12-04 00:29 mhall
Added an option for using a user-supplied threshold value
2007-12-01 03:57 fracpete
initModel is now also called in setModelFile(File) and getCapabilities(), otherwise the capabilities cannot be determined correctly:
https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2007-December/011783.html
2007-11-28 03:04 mhall
Initial import of Martin's feature selection stuff.
2007-11-27 03:30 mhall
Added Martin Guetlein's bug fix for merit caching
2007-11-23 02:44 mhall
Fixed a potential division by zero in out-of-bag routine
2007-11-21 22:37 fracpete
added capability MISSING_CLASS_VALUES (kernels don't care about class)
2007-10-30 03:27 mhall
When saving trained model to binary file now saves training header (like the Explorer does). Similarly, when loading binary model attempts to load the header from the stream as well.
2007-10-29 20:34 fracpete
support of an attribute range to base the distance calculation on is now mandatory for all distance functions
2007-10-28 23:41 fracpete
new distance functions (Manhattan and Chebyshev) which share the same abstract superclass as the Euclidean distance function: NormalizableDistance
2007-10-28 23:41 fracpete
branches:  1.1.2;
new distance functions (Manhattan and Chebyshev) which share the same abstract superclass as the Euclidean distance function: NormalizableDistance
2007-10-28 23:31 fracpete
added scale and transformation parameter to get different range output than [0,1]
2007-10-25 01:59 fracpete
branches:  1.3.2;
PCA filter based on the attribute selection evaluator weka.attributeSelection.PrincipalComponents
2007-10-09 23:08 mhall
Now throws an exception if NaN occurs during standardization. This avoids introducing missing values into the filtered data set in this situation.
2007-10-06 22:44 fracpete
removed "final" from getOutputFormat method declaration since that caused VerifyError exceptions in derived filters on Mac Books:
https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2007-October/011339.html
2007-09-20 21:43 fracpete
test for missing values missed the last column (in case it wasn't the class attribute!)
2007-09-10 21:53 fracpete
mathematical expression got only evaluated for toString/getMeasure, but for actual setup of filter/classifier. Didn't pop up as error since most Weka schemes do NOT check the input in the get-method (you only get wonky results). Was discovered by M. Fatih Akay using LibSVM, since LibSVM threw an exception in case of invalid parameters, but the values would have been valid IF the expression would have been evaluated.
2007-09-07 07:05 fracpete
classifyInstance returned the probability instead of the index from the class probability array for nominal classes (in case of average/product/majority/min/max rule). Thanks to Samuel Robert Reid
2007-09-07 07:02 fracpete
tokenizer could not be serialized due to unserializable StringTokenizer class. changed m_Tokenizer to transient.
2007-09-03 22:11 fracpete
removed classifyInstance method, since there were discrepancies between the labels returned by classifyInstance and the best one returned by distributionForInstance (svm_predict returned other labels than svm_predict_probability???)
used version of libsvm: 2.84
2007-08-30 23:11 mhall
Fixed bug that prevented quoted ? from being parsed as a label rather than a missing value
2007-08-26 23:08 mhall
Fixed inconsistent naming of toolTipText method
2007-08-23 02:25 mhall
Fixed bug that caused stack overflow by removing all uses of Utils.sm and Utils.gr
2007-08-22 02:00 mhall
Initial import. Contributed by Geoff Webb's group.
2007-08-20 00:58 fracpete
stopwords no longer get ignored
2007-08-09 22:36 fracpete
savers were too restrictive in writeBatch method: instead ALWAYS having a file to write to, one needs only either a file OR a stream (both are used through a Writer)
2007-08-01 04:06 fracpete
added general Jython support
Jython classes that are derived from weka.classifiers.Classifier can be executed with the wrapper classifier weka.classifiers.JythonClassifier (serializing/deserializing a trained model is unfortunately NOT possible, a call of buildClassifier is always necessary after deserialization)
2007-08-01 04:06 fracpete
branches:  1.1.2;
added general Jython support
Jython classes that are derived from weka.classifiers.Classifier can be executed with the wrapper classifier weka.classifiers.JythonClassifier (serializing/deserializing a trained model is unfortunately NOT possible, a call of buildClassifier is always necessary after deserialization)
2007-07-15 21:36 mhall
Instance variable holding an Evaluation object was not set to null after buildClassifier, resulting in not serializable exception in Experimenter
2007-07-11 21:52 fracpete
added the "output of additional attributes" option to the "more options" dialog in order to output attribute values alongside the predictions (just like on the commandline with "-p")
2007-07-09 02:41 cvs_rrb
Fix related to Markov blanket correction
2007-07-09 02:41 cvs_rrb
Fix related to Markov blanket correction
2007-07-09 02:38 cvs_rrb
Markov blanket correction fix
2007-07-03 23:48 mhall
Instance weights were being cast to ints for some reason. Removed the casts
2007-07-03 01:24 mhall
Fixed bug in kthSmallestValue that prevented the largest value from being returned
2007-07-02 04:10 cvs_rrb
Access to nodes (usefull for Bayes network GUI)
2007-07-02 03:59 cvs_rrb
Initialize from Editable Bayes network
2007-07-02 03:58 cvs_rrb
Access from Bayes network GUI
2007-07-02 03:58 cvs_rrb
Bayes network algorithm for calculating marginal distributions
2007-07-02 03:54 cvs_rrb
Icons for Bayes netowrk GUI
2007-07-02 03:54 cvs_rrb
Icons for Bayes netowrk GUI
2007-07-02 03:51 cvs_rrb
Bayes network GUI
2007-07-02 03:49 cvs_rrb
access for Bayes net GUI
2007-07-02 03:45 cvs_rrb
Access for Bayes Net GUI
2007-07-02 03:44 cvs_rrb
Access for EditableBayesNet
2007-07-01 22:42 mhall
Sebastian has given copyright to UoW
2007-06-29 04:00 fracpete
disabled EnsembleLibrary for now due to bugs: craps out if the GOE-editable-object depth is more than 2, e.g., a FilteredClassifier with SMO (the kernel is the third level) generates a StackOverFlowError
2007-06-29 03:42 fracpete
disabled EnsembleLibrary for now due to bugs: craps out if the GOE-editable-object depth is more than 2, e.g., a FilteredClassifier with SMO (the kernel is the third level) generates a StackOverFlowError
2007-06-26 21:05 fracpete
fixed technical information (booktitle -> journal)
2007-06-26 04:19 mhall
branches:  1.1.2;
file movePublicDomain.sh was initially added on branch waikato.
2007-06-25 22:50 mhall
This class has been placed into the public domain by the author, Werner Randelshofer
2007-06-25 02:05 fracpete
de-serializing savers no longer sets the file, since the KnowledgeFlow handles that itself, by using prefix, directory and relation-name for the filename
2007-06-25 02:02 fracpete
statement variables are now transient (otherwise an exception is raised that these objects aren't serializable)
added checks whether a DB connection is active (after de-serializing for instance)
2007-06-25 02:01 fracpete
added checks whether a DB connection is active (after de-serializing for instance)
2007-06-25 01:59 fracpete
doesn't stumble over '' anymore by replacing it with one single quote (creates an invalid SQL statement)
2007-06-25 01:57 fracpete
removed restriction that nominal attributes have to have at least one attribute. ARFF files generated from databases can have empty nominal attributes, e.g., "Summary" from a results table of the Experimenter may not be filled in (only missing values). Such a table could not be loaded again in the Explorer again.
2007-06-22 04:35 mhall
UoW has the copyright.
2007-06-22 01:25 fracpete
added information about how ROC and AUC are computed, based on this Wekalist post:
https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2007-June/010386.html
2007-06-21 23:24 mhall
Changed copyright to University of Waikato as Len gave over the copyright.
2007-06-21 23:13 mhall
Changed copyright notice as Alex has given copyright to the University
2007-06-21 22:54 mhall
branches:  1.3.2;
Changed copyright back to Haijian as he was not on a scholarship when this code was written.
2007-06-21 22:17 fracpete
added command "history" to list the current conent of the command history
last X number of commands are saved in $HOME/SimpleCLI.props (X = HistorySize in props file)
2007-06-20 23:02 fracpete
instead of listing only number of test instances, the split evaluators now list number of training and test instances (like the ClassifierSplitEvaluator)
2007-06-20 04:12 fracpete
one Utils.quote call too much
2007-06-20 01:39 fracpete
source code generation now generates a static Weka filter
2007-06-19 04:28 fracpete
fixed "sourcing" of classifiers:
- Explorer outputs now the complete class, not just the static class, but also the wrapper class
- wrapper class has now a main method and a toString method to enable calls from the commandline like with any other classifier
- reversed all "public class" to "class" again
- class attribute now gets set to missing explicitly before the static class is called ("classify")
- updated weka.classifiers.CheckSource to reflect changes (source code has to be derived from weka.classifiers.Classifier now)
- fixed OneR's toSource method
2007-06-19 00:52 fracpete
added new interface weka.filters.Sourcable and let a few simple filters implement it to generate source code (option "-z classname")
2007-06-19 00:46 fracpete
fixed typo in Javadoc
2007-06-18 21:07 fracpete
fixed toString() method if no model built yet
2007-06-18 01:54 fracpete
fixed tilde in BibTeX Javadoc output
2007-06-18 01:54 fracpete
branches:  1.3.2;
fixed tilde in BibTeX Javadoc output
2007-06-18 01:54 fracpete
fixed tilde in BibTeX Javadoc output
2007-06-18 01:54 fracpete
branches:  1.65.2;
fixed tilde in BibTeX Javadoc output
2007-06-18 01:54 fracpete
branches:  1.4.2;
fixed tilde in BibTeX Javadoc output
2007-06-18 01:54 fracpete
branches:  1.3.2;
fixed tilde in BibTeX Javadoc output
2007-06-18 01:54 fracpete
branches:  1.6.2;
fixed tilde in BibTeX Javadoc output
2007-06-18 01:53 fracpete
fixed preamble comment indentation
2007-06-18 01:51 fracpete
branches:  1.5.2;
fixed Javadoc of -C <cache size> option
2007-06-18 01:49 fracpete
branches:  1.6.2;
updated Javadoc of -C <cache size> option of SMO
2007-06-18 01:47 fracpete
branches:  1.13.2;
fixed Javadoc of -C <cache size> option
2007-06-18 01:44 fracpete
reversed the escaping of the underscore (only tilde now)
2007-06-17 03:54 fracpete
added additional output to the log with setup for the AttributeSelectedClassifier
2007-06-15 06:40 fracpete
added three new measures to the Experimenter:
  Serialized_Model_Size - size of the serialized classifier object
  Serialized_Train_Set_Size - size of the serialized train set
  Serialized_Test_Set_Size - size of the serialized test set
2007-06-14 02:09 fracpete
fixed serialVersionUID
2007-06-14 01:51 fracpete
method "mapClasses" is now public and static instead of private
2007-06-08 01:35 fracpete
incorporated changes between Jama 1.0.1 and 1.0.2 in Weka's classes
see also Jama's changelog: http://math.nist.gov/javanumerics/jama/ChangeLog
2007-06-06 04:24 fracpete
minimum number of instances of capabilities is now set to number of bins
2007-06-05 21:06 fracpete
on Macs the updateCurrentConverter() method crashes with a NullPointerException if the setAcceptAllFileFilterUsed(boolean) method is used before the dialog is even displayed (e.g., in the showSaveDialog(Component) method). Looks like Java 1.5 fires property changes differently on Macs (Win32 and Linux works fine). Thanks to Craig Bruce.
2007-06-01 04:32 fracpete
updated version