rsubgroup NEWS

For a detailed list of changes, see the svn commit history at
svn://r-forge.r-project.org/svnroot/subgroup/

Changes in version 1.1 (2021-02-22)
===================================
* provide automatic discretization options of VIKAMINE kernel for
  numeric attributes, they are internally discretized
  SDTaskConfig provides new options.
  defaults: discretize = TRUE, nbins = 3
* Improved/extended tests for this.


Changes in version 1.0 (2020-04-20)
===================================
* internal enhancements in subgroup.jar, i.e., the VIKAMINE kernel library,
  e.g. according to better error messages relating to the R connection.
* Improved documentation and examples.


Changes in version 0.9 (2020-03-04)
===================================
* internal enhancements in subgroup.jar, i.e., the VIKAMINE kernel library
* rsubgroup requires >= Java 8 (i.e., >= java.version 1.8)   


Changes in version 0.8 (2019-05-23)
===================================
* Improvements
  * included new SDTaskConfig#parfilter: Provides the minimal improvement
    value for the postfilter (for min-improve-* filters), or the applied
    significance level (P) for sig-improve-* filters.
  * updated org.vikamine.kernel version
  * package-internal: fixed Java requirements (string) in DESCRIPTION

* Bug fixes.
  * fixed bug in automatic discretization used in rsubgroup VIKAMINE kernel


Changes in version 0.7 (2015-07-01)
===================================
* Improvements
  * document setting Java heap space before loading the rsubgroup library.
  * Improve error handling (exception signaling) when running subgroup discovery
  using an ARFF file directly.
  * SDTaskConfig now provides an option mintp, that allows to set the minimal
  true positives threshold to be contained in a subgroup, which is usually
  very effective for pruning.
  * The Pattern class now contains a list of selection expressions (selectors)
  for the subgroup, not only the description. Using the is.pattern.matching
  function, a match of a pattern and a data instance can be checked now.
  * In SDTaskConfig, postfilter can be a single filter or a vector of filters,
  that are then applied in order on the results. This allows e.g., the combination
  of minimal improvement filtering with weighted covering post-processing.
  * Implement/enable new quality function (Adjusted residuals, cf. Agresti 2007)
  ==> qf="ares"
  * For a binary target variable, the resulting patterns now also store the
  chi-squared value comparing subgroup and population w.r.t. the target in the
  parameters field.
  * ToDataFrame shows the chi-squared value for a binary target.

* Bug fixes:
  * fix providing attributes=NULL (i.e., automatically include all attributes)
  into subgroup discovery
  * fix max-attribute-value bug in SGSelectorGenerator, causing the inclusion
  of two few selectors in subgroup discovery methods


Changes in version 0.6 (2014-09-10)
===================================
* Improvements:
  * optimizations in the beam search algorithm
  * significant memory optimization (dataset storage, access)
  * ARFF: enable import of "empty" attributes, i.e., with an empty value domain
  (this can occur, for example, when importing columns with only 'NA' in R)  

* Bug fixes:
  * exclude target from subgroup attributes
  * fix loading of ARFF file, for attributes/values with trailing spaces