Kaadugal is a parallelized multi-core C++ implementation of the random forests algorithm for classification, regression, and structured prediction problems. Kaadugal is optimized for performance on multi-core systems with tens of processors and a shared memory.
Kaadugal is known to work on multiple platforms including Linux and Windows. It uses C++11 features and therefore requires a compiler that supports these features. It also uses OpenMP, so please make sure it is available. Otherwise no speedup will be possible.
Kaadugal is the English phoneticization of the word காடுகள் (forests) in the Tamil language.
[15-Oct-2014]: Classification forests are done! See below for usage example.
[1-Oct-2014]: This library isn't fully implemented yet. I am actively working on it though and hope to finish soon.
Kaadugal is a header only library that can simply be included in your projects. The kaadugal/include path should be in your project include path.
The examples directory contains a toy problem for training and testing simple classification and structured prediction problems. The steps to build the examples are:
$ pwd <SOME_DIR>/kaadugal $ mkdir build && cd build $ cmake ../examples/ $ make
To run the example classification problem first learn the forest:
./classify train ../examples/config/<CONFIG_FILE> <OUTPUT_FOREST_PATH> ../examples/data/<DATA_FILE>
Then test the forest on input data using
./classify test <INPUT_FOREST_PATH> ../examples/data/<DATA_FILE>
The util directory contains some utilities for manipulating learned forests/trees, compressing, etc. Follow similar steps as examples to build these utilities (replacing examples with util).
Srinath Sridhar (firstname.lastname@example.org) Max Planck Institute for Informatics