Log of RL Changes

• Attribute Editing is now improved. Also, drag and drop has been enabled for value hierarchies.
• You can sort the attributes by column header. There's no way to get back to the original sort order--does that matter?
• All attribute order code has been shifted to the AttributeDlg class. AttributeList doesn't need to know that information since attribute ordering doesn't affect SAL.

• Class DataKeeper no longer exists.
  It's gone. Zap. Import functionality was switched to a new class, rl.io.ImportDataDelim. The user can specify any single character to serve as the delimiter, but the default is a tab. The statistics methods were pushed directly into DataList. The various data lists are now stored in ProgramState directly. All references to DataKeeper should be gone (except possibly in comments).
• Fixed Bug 96. RL no longer produces an empty table when you perform a new search using the beginner tab.
• Fixes Bug 97. Empty lines are ignored by RL.
• Closed Bug 75. The resources (including help files) are accessible through the GUI, if the resources directory is archived as part of the RL executable JAR file, or is in the same location as the JAR file. Should the resources be JARred with the RL executable?

• Merged class AttributeKeeper into class AttributeList
• Can now import attribute definitions from file
• Can now import value hierarchies from file
• Much work on storing value hierarchy
• Work on importing rules, including hard design questions (everyone)
• Work on separating the UI from finctionality and IO from everything else

• Renamed file attribute/Conjunc.java to attribute/Conjunct.java0 since this file was causing the whole applicatio to fail to compile.
• MacOS now runs JDL1.4 and therefore RL.

• Fixed bug with secondary criterion for ranking rules - the criterion was primary.
• Removed call att.primeMatrix() from MainFrame.
• Added a call attList.primeMatrrix() in SAL.
• Removed files images/Thumbs.db, JavaHelpSearch, Map.jhm, .nbattrs, RLHelp, RLHelp.hs, RLHelpIndex.xml, RLHelpTOC.xml from resources/.
• Added commented code in HNode.
• Added new trivial class attribute.Conjunc.
• Removed @deprecated tag in HeapPriorityQueue.
• Reformatted Predictor, RuleComparator, RuleLearnerDataComparator

• Changed the secondary critrerion for ranking rules fron p-value to TruePositive coverage. This impacts eg.SingleBest.

  Normal
      cluster-1
      cluster-5
      cluster-6
  Cluster
      cluster-2
      cluster-3
      cluster-4
  

• Import missclassification costs from file (Eric a long time ago)

• Fixed the learning bug that caused RL version 2002-08-06 and later to learn different (and less accurate) rules than RL version 2002-08-02. The bug was caused by the statement "return false" in SAL.isBadRule(), the code that tests coverage (line 767).
• Fixed the matching bug that caused RL to produce different accuracy and coverage than version 2002-11-06 during inference. The bug was in Conjunction.matchDatum(), and was fixed by negating the If Condition (line 54) and return statements (lines 55 and 58).

• Fixed NullPointerException when train or test file has empty lines.
• Fixed In the Info Panel when an attribute's use is changed from Output to something else, the info gain column should be cleared.

• Fixed Won't start: Can't find main class. [Probably because there were compile errors in the source, which compiled and ran on Linux.]
• Attrubute Panel now calculates the attribute information gains automatically as soon as a target attribute is set.
• The Info Gain button is renamed to Sort By Info Gain, and sorts the rows of the table by decreasing information gain.
• Class DataItem is renamed to Datum.

• The Data table in the output now shows matching rules in columns by RHS.
• The Rules table in the output now shows predicted and matched data in columns by observed target value.
• Attribute dialog box now has "Done" and "Edit..." buttons.
• No more "can't export empty heap to ArrayList" exceptions on the command line prompt.
• Re-ordered the expert panel.
• Fixed After importing data, when I go to the attribute definition frame, pressing "information gain" doesn't work. It works after you learn some rules, but not before. [A bug from v0.83 (2002-08-06) or earlier.]
  
• Fixed Rule 3 has 0 items in "predicted" 7 in "matched", 15 TPs and CF of almost 600. [I still can't replicate that.] Seems it's not a bug -- the computation was based on signal-to-noise ratio, which was not set.
• Fixed In the attributes dialog, editing the "number of uses" field failed.
  
• Fixed Can't edit the value hierarchy of attributes from the Attribute dialog. I think the problem was limited to Linux.
  
• Lots of code cleanup.

• Cleaned up html in output
  
• Beautified search Panel
  
• Beautified Data Table in output
  
• Added to Data Table in output
  
• New Rules Table in output
  
• Cleaned up code and comments in various files including eg/*.java

• Minimum coverage and minimum positive coverage are now consistent - both interpret values less than 1 as percentages.
• Importing new training data no longer resets the EG method, CF method, and thresholds.
• There is an additional import option - attribute defintion file. Create that file as a tab-delimited text file with the following format.

      attrib1 attrib2 attrib3  ...  attribN
      tag1    tag2    tag2     ...  tagN
      type1   type2   type3    ...  typeN
      use1    use2    use3     ...  useN
  

  tag is a semantic tag for use in constraints. Use none as a filler if needed.
  type is either discrete, continuous, or set
use is either input, output, identification, or ignore

  Attribute type information will override automatic type detection. Be sure you know what you're doing.

• Feature added: cost-based learning - works as follows
  There is an additional import option - misclassification cost file. Create that file as a tab-delimited text file with the following setup target_class predicted_value actual_value cost_of_error cost_of_error is a floating point value that represents the penalty incurred for making that prediction mistake. The default penalty is 1.0. This is to avoid divide-by-zero errors when calculating the CF to average cost ratio. All of the evidence gathering methods that normally select based on CF will now do so based on that ratio. Right now, I can only guarantee correct functionality for discrete target classes. Be certain that discrete targets are not numerical (eg turn 4 into G4 or four).
• Feature added: All-pairs evaluation evidence gathering method.
  There is now an EG method previously only used for my testing. It uses all of the other methods and compares them. The comparision results can only be seen if running the program from within an IDE or from the command line.

• Fixed: All the evidence gathering methods work the way they're supposed to (Don't ask).
• Changed: After running comparisons of all the evidence gathering methods, weighted voting is now the default. Also some methods have been removed due to poor performance.

• Fixed: I believe that the choking problem has been fixed, or at the very least reduced. I ran a 146-attribute, ~2300-case dataset and it didn't crash like it would with the 03-26-02 code. It still took a very long time, though, so one of my goals for the next version is to make the code more efficient.
• Changed: I am using a new version numbering system. I estimate that JavaRL is about 80% complete, so this version is 0.8. The old numbering system remains for previous versions. To avoid ambiguity, all old version numbers say "(old numbering)" after them. When the new version 1.0 comes along, no mention will be made of the old system.
• Changed: The Beam class (Beam.java) is now deprecated. I have created a PriorityQueue interface. The only implementation thus far is a binary heap (HeapPriorityQueue.java). The SAL code (SAL.java) as well the evidence gathering code had to be modified to accomodate the new data structure. Several private classes were made public in separate files. SAL.java was just to large and unwieldy to have so many classes stored within it. Besides, it's better coding practice to put classes into seaprate files.
• Feature added: new evidence gathering method - "Most specific AND highest CF rule" (MostSpecificSingleBest.java). From several "best" rules, it picks the one with the most conjuncts AND the highest CF. P-value is used as a tie-breaker. Note that this is different from SingleBestSpecific.

Other comments & recommendations

• I suggest using Borland JBuilder 6 Professional ($100 through Pitt Book Center) as a development environment. It's a lot faster, more stable, and intuitive than Forte.
• The changes I have made and will make in the future are drawn from the Needs/wants list at http://www.cs.pitt.edu/~edwst7/JavaRL/JavaRL_bugs.txt.

• P-values are now correctly calculated as the right-tail The p-values are calculated by a child of CertaintyFactor, but are not included in the CF list. There are notes in CertaintyFactor.java explaining why.

          (Rule.java)
          (CertaintyFactor.java)
          (PValueRight.java)
          (PValue.java)
  

• Larger datasets are now usable.

          (SAL.java)
  

• Fixed class-checks in equals method of the attribute value nodes.

          (DiscreteNode.java)
          (IntervalNode.java)
          (SetNode.java)
  

• Minor bug fixes when I came across them.
  ExpertPanel defaults are now consistent Information gain is calculated for the results w/o user action Error messages if the user tries to specify multiple target/id fields
• Replaced String concatenation with StringBuffer.append when appropriate
• Fixed misspellings of continuous and hierarchy
• Attempted to rearrange comments based on Java conventions
• Attempted to reorganize/rename files based on Java conventions
• Some classes had implemented the serializable interface, I removed this implementation in many cases.
• Added the ability to specify an attribute for ID.
  There is currently no check for uniqueness of the ID field There is no automatic creation of IDs if a field is not specified
• Added the date and file name to exported results and rule files
• Added attribute definitions to the exported rules file.
  This only contains attributes used in the rules for various reasons
• Changed default file extensions for exported files
• The value-hierarchy editor has been updated to be more useful