What is it?
CORELS is a custom discrete optimization technique for building rulelists over a categorical feature space. The algorithm provides theoptimal solution with a certificate of optimality. By leveragingalgorithmic bounds, efficient data structures, and computational reuse,it achieves several orders of magnitude speedup in time and a massivereduction of memory consumption. This approach produces optimal rulelists on practical problems in seconds, and offers a novel alternativeto CART and other decision tree methods.
See theC++implementation, thelivewebsite, thePythonimplementation, thearXivpaper, theJMLRpaper, theSysMLpaper, thesenior thesis ortheKDD2017 paper for more.
More about Corels can also be read inthis recent post atThe Morning Paper.
Illustration
With thanks to thePython implementationfor the image.
What is this package?
We useRcpp to connecttheCorels C++implementation to R.
Status
Installs and works fine, and passedR CMD check. Severalextensions possible, see below.
Installation
As the package is not (yet?) on CRAN, do
remotes::install_github("eddelbuettel/rcppcorels")Note of theGNU GMP library is nowoptional;configure will enable (via a-DGMPdefine and link instructions) if found. GMP will improve performance, soyou may want to dosudo apt-get install libgmp-dev, orwhatever equivalent command you need to install it on your system.
TODOs
Plenty such as adding Travis CI support, adding configure code todetect GNU GMP presence, adding examples, factoring out (input) datareader code, possibly visualizing decision trees, and more.
Author
Dirk Eddelbuettel wrote the R package and integration.
Nicholas Larus-Stone and Elaine Angelino wrote the C++ implementationof Corels.
Elaine Angelino, Nicholas Larus-Stone, Daniel Alabi, Margo Seltzer,and Cynthia Rudin wrote the paper.
Corels uses the rulelib library by Yanget al described inthe2016 arXiv paper byHongyu Yang, Cynthia Rudin, and Margo Seltzer withthis code repo and in the2015 arXiv paper by BenjaminLetham, Cynthia Rudin, Tyler H. McCormick and David Madigan nowpublished inAnnals of Statistics.
License
This package is released under the GPL-3, as is Corels.
The rulelib library is released under the MIT license.
Initially created: Sun Nov 03 10:03:02 CST 2019
Last modified: Sun May 26 09:57:46 CDT 2024