bigrf News

For a detailed list of changes, see git commit history at
https://github.com/aloysius-lim/bigrf.


Changes in version 0.1-11 (16th May 2014)

Package updates:

  * Removed BH package as an Imports dependency; it is now only in LinkingTo.


Changes in version 0.1-10 (14th May 2014)

Improvements:

  * grow() and predict() methods now print the summary at the end only if
    trace >= 1.


Changes in version 0.1-9 (20th February 2014)

Package updates:

  * Reduced number of trees used in examples, to reduce running time.

  * Removed outdated information from DESCRIPTION file.


Changes in version 0.1-8 (21st September 2013)

Package updates:

  * DESCRIPTION file updates to comply with package submission requirements.

  * Removed INSTALL-WINDOWS.


Changes in version 0.1-7 (20th September 2013)

Package updates:

  * Added README.md (used on github only) to .Rbuildignore.


Changes in version 0.1-6 (28th June 2013)

Bug Fixes:

  * Fixed segfault when running bigrfc on categorical-only data (i.e. data with
    no numeric attributes).


Changes in version 0.1-5 (11th April 2013)

Package updates:

  * Added LinkingTo dependency BH, due to changes in dependencies in the
    bigmemory package.

Bug Fixes:

  * Predict method now observes the cachepath parameter.


Changes in version 0.1-4 (5th Mar 2013)

Package Updates:

  * Fixed license information.

Improvements:

  * Added check for R user interrupt after processing each tree node.

  * Added additional level of verbosity for proximities and scaling methods.

  * Added additional stop conditions in scaling method, to prevent infinite
    loops. Scaling computation now stops after 1,000 iterations per co-ordinate,
    when intermediate variable ynorm does not decrease by more than 0.05%, or
    when ynorm crosses below a certain threshold determined during the
    iteration.

Performance Enhancements:

  * Better memory management in grow method, releasing allocated memory after
    finding the best split for each node. This should reduce RAM usage overall.

  * More efficient algorithm for proximities method. Reduced number of passes
    through trainprednode in each tree from forest@ntrees * forest@nexamples to
    forest@ntrees * [sum over all trees(tree@nnodes)].

  * Parallelised proximities method.

  * Parallelised some parts of scaling method.


Changes in version 0.1-3 (18th Feb 2013)

Package Updates:

  * Suggests the doParallel that uses the parallel package that ships with R,
    instead of doMC, doMPI, doRedis, doSNOW.

  * Requires R >= 2.14, when the parallel package was released.


Changes in version 0.1-2 (17th Feb 2013)

Bug Fixes:

  * Set the OS type for this package to unix only, as dependency bigmemory
    is now only available for unix. However, Windows users can still use this
    package by manually downloading and installing bigmemory (windows binary for
    older version available on CRAN) and bigrf (install from source).

  * Added file .gitignore to .Rbuildignore, to pass R CMD check.


Changes in version 0.1-1 (15th Feb 2013)

  * Initial release
