Planned changes in next versions:

* combine function that allows parallel computation
* more refactoring to reduce the memory footprint (store indices instead subsets)
* stratified sampling when creating row partitions to avoid problems when too many row partitions
* check for constants in subsets

---------------------------------------------------------------
Change log:
* Version 0.3.0: Sep 2015
-added predict.all parameter
-fixed a bug related to the filter parameter

* Version 0.2.2: May 2015
-improved oversampling in balanced cases
-automatically remove constants and/ or near constants (filter parameter)

* Version 0.2.1: May 2015
-stratified sampling when making training and test set
-parameter for oversampling to avoid problems related to subsetting
-added oversampling

* Version 0.2.0: March 2014
-fix print when loading package
-allow parameters for genetic algorithm and random forest
-Refactoring: improved code readability and reduced complexity to improve maintainability
-switched to roxygen for documentation
-replaced package ROCR with AUC for faster evaluation of objective function
-updated reference

* Version 0.1.2: May 2013
-Fixed a bug that could occur in very rare cases

* Version 0.1.1: February 2013
-Fixed a bug that occured for data frames with only numeric features

* Version 0.1.0: November 2012
-Package submitted to CRAN with two functions: kernelFactory and predict.kernelFactory