en:Bugs (3.5.7)

From WekaDoc

This article lists known bugs in this release of Weka. This is not used as a bug-tracker, only as information. If you find a bug, please report it on the Wekalist.

Table of contents

1 GUI 1.1 Explorer 1.2 Experimenter 1.3 KnowledgeFlow 1.4 SimpleCLI 1.5 ArffViewer 1.6 SqlViewer 1.7 EnsembleLibrary 2 Command line 2.1 Associators 2.2 Attribute selection 2.3 Classifiers 2.3.1 weka.classifiers.meta.ThresholdSelector 2.4 Clusterers 2.5 Data generators 2.6 Experiments 2.7 Filters 2.7.1 weka.filters.supervised.attribute.AddClassification 2.7.2 weka.filters.supervised.instance.Resample 2.7.3 weka.filters.unsupervised.attribute.AddCluster 2.7.4 weka.filters.unsupervised.attribute.ClassAssigner 3 Other 3.1 weka.core.Capabilities 3.2 weka.core.stemmers.SnowballStemmer

GUI

Explorer

Experimenter

KnowledgeFlow

SimpleCLI

ArffViewer

SqlViewer

EnsembleLibrary

Command line

Associators

Attribute selection

Classifiers

weka.classifiers.meta.ThresholdSelector

• option handling of -C parameter was incorrect: initial value for m_ClassMode was OPTIMIZE_POS_NAME, default in setOptions was OPTIMIZE_LFREQ instead, getOptions returned m_DesignatedClass instead of m_ClassMode
  • 08/04/2008 - found and fixed by FracPete

Clusterers

Data generators

Experiments

Filters

weka.filters.supervised.attribute.AddClassification

• AddClassification filter wasn't copying the values of string and relational attributes to the output, resulting in NullPointerExceptions
  • 28/03/2008 - found (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2008-March/012807.html) by Fernando Cela Diaz
  • 30/03/2008 - fixed (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2008-March/012829.html) by FracPete

weka.filters.supervised.instance.Resample

• Resample filter crapped out if using 100% sample size and -no-replacement option: as soon as one of the vectors holding the indices for a class became empty, the Random.nextInt call was given an upper limit of 0, which resulted in an exception that n has to be positive.
  • 22/01/2008 - found by Daniel Rodriguez
  • 22/01/2008 - fixed by FracPete

weka.filters.unsupervised.attribute.AddCluster

• The filter generated a error message "problem filtering instances:null". This was due to an exception thrown in case an instance couldn't be clustered. Exceptions are now caught and a missing value is set as cluster.
  • 25/03/2008 - found (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2008-March/012759.html) by Rafiqul Haque
  • 30/03/2008 - fixed (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2008-March/012830.html) by FracPete

weka.filters.unsupervised.attribute.ClassAssigner

• the method setClassIndex didn't not subtract 1 from the index after parsing the integer value (internal index is 0-based, whereas the one returned is 1-based).
  • 16/01/2008 - found and fixed by FracPete

Other

weka.core.Capabilities

• getAttributeCapabilities()/getClassCapabilities() were not necesssarily returning the correct capabilities due to the use of Capabilities.enable(...) instead of just filling the HashSet. The enable method automatically enables encompassed capabilties, i.e., if one enables NOMINAL_ATTRIBUTES, then UNARY/EMPTY will be enabled as well. Led to display discrepancies in the Explorer's Capabilities dialog.
  • 31/12/2007 - found (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2007-December/011966.html) by Martin Řehánek
  • 31/12/2007 - fixed (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2007-December/011973.html) by FracPete

weka.core.stemmers.SnowballStemmer

• method objects for calling the snowball library weren't re-initialized after de-serialization and generated NullPointerExceptions. Just running the stemmer as part of the StringToWordVector from commandline was fine though.
  • 12/03/2008 - found and fixed (https://list.scms.waikato.ac.nz/mailman/htdig/wekalist/2008-March/012593.html) by FracPete