GARF - A Regression Random Forest Library

Version 0.1

GARF is a C++ library for training and testing Regression Random Forest
models, popularised in Machine Learning and Computer Vision since their
invention in 2001 by Leo Breiman. Various implementations are available
(OpenCV, Scikit-learn) but for reasons including wanting more control over
the types of features that can be chosen, wanting the highest performance
possible, and just generally wanting to get my hands dirty, I wrote my own.

I've been using this in my research for nearly the past year, and while I
wouldn't claim it is anything other than alpha software, it's worked pretty
reliably for me.

Features:
* Parallelisation with Intel Thread Building Blocks
* Serialization (save and load trained forests) with Boost.Serialize
* Python interface with Boost.Python

Requirements:
* C++ 11 compiler. I generate my random numbers using the new C++ 11
  classes so this is a hard requirement. I use clang++ 4.2 with libc++
  on OS X 10.8.3 - I imagine it will be mediumly easy to compile this
  code on Linux, mediumly hard on Windows.
* Intel Thread Building Blocks - I used 4.1u3.
  https://www.threadingbuildingblocks.org/
* Eigen headers (required internally for all vectors & matrices).
  I used version 3.2.0
* Boost for smart pointers, Python, serialisation & parallelisation.
  I used 1.54.0 - note that the compiled parts (such as Python)
  need to be compiled in C++11 mode or you are liable to get weird
  linking areas. If you're using Homebrew on a Mac to install boost 
  make sure you use the --with-c++11 flag.
* Google Test API if you want to compile the included (admittedly not
  that extensive) unit tests
* Python 2.7 (previous versions may work..) with Numpy for the python
  integration

To use GARF as a C++ library use the Makefile and tests/forest_test.cpp
as guidelines. Better documentations and examples TBC!


Python instructions:

1. Change paths if necessary in python_bindings/setup.py to point to where your
   Boost, TBB and Eigen are installed. Eigen only needs to be
   in the include directories, Boost and TBB have to be in the include
   and library directories.
2. From a terminal inside python_bindings/setup.py
	$ python setup.py build
	$ python setup.py install
   You may need sudo for the instal command, depending on your python setup.

3. See tests/python_example.py for a simple example of usage.


This code will be hosted for the foreseeable at https://github.com/malcolmreynolds/garf
- should you use it for anything I would love to hear about it. Patches / PRs
even more welcome.