Regularized Greedy Forest

Regularized Greedy Forest wrapper of the 'Regularized Greedy Forest' <> 'python' package, which also includes a Multi-core implementation (FastRGF) <>.

UPDATE 17-08-2018: Now RGF is part of the official RGF repository and active development is performed there. The mlampros/RGF repository is archived.

The RGF package is a wrapper of the Regularized Greedy Forest (RGF) python package, which also includes a Multi-core implementation (FastRGF). More details on the functionality of the RGF package can be found in the blog-post and in the package Documentation.

UPDATE 26-07-2018: A Singularity image file is available in case that someone intends to run RGF on Ubuntu Linux (locally or in a cloud instance) with all package requirements pre-installed. This allows the user to utilize the RGF package without having to spend time on the installation process.


Rie Johnson and Tong Zhang, Learning Nonlinear Functions Using Regularized Greedy Forest




System Requirements

  • Python (2.7 or >= 3.4)

All modules should be installed in the default python configuration (the configuration that the R-session displays as default), otherwise errors will occur during the RGF package installation (reticulate::py_discover_config() might be useful here).

Debian/Ubuntu/Fedora [ installed / tested on Python 2.7 ]

First install / upgrade the dependencies,

sudo pip install --upgrade pip setuptools
sudo pip install -U numpy
sudo pip install --upgrade scipy
sudo pip install -U scikit-learn

Then, download the rgf-python package and install it using the following commands,

git clone
cd rgf/python-package
sudo python install

FastRGF will be installed successfully only if gcc >= 5.0.

Mac OS X [ installed / tested on Python >= 3.4 ]

Upgrade python to version 3 using,

brew upgrade python

Then install the dependencies for RGF and FastRGF

sudo pip3 install --upgrade setuptools
sudo pip3 install -U numpy
sudo pip3 install --upgrade scipy
sudo pip3 install -U scikit-learn

The FastRGF module requires a gcc >= 8.0. To install gcc-8 (or the most recent gcc) with brew follow the next steps,

# before running the following commands make sure that the most recent Apple command line tools for Xcode are installed
brew update
brew upgrade
brew info gcc
brew install gcc
brew cleanup

After the newest gcc version is installed the user should navigate to /usr/local/bin and if a gcc file exists (symbolic link) then the user should delete it. Then the user should run the following command,

sudo ln -s /usr/local/bin/gcc-8 /usr/local/bin/gcc

The user should then verify that the gcc has been updated using,

gcc -v
which gcc

After the new gcc is installed the user should continue with the installation of rgf_python,

git clone
cd /rgf/RGF/build
export CXX=/usr/local/bin/g++-8 && export CC=/usr/local/bin/gcc-8
cmake /rgf/RGF /rgf/FastRGF
sudo make install
cd /rgf/python-package
sudo python3 install

After a successful rgf-python installation the user should open an R session and give the following reticulate command to change to the relevant (brew-python) directory (otherwise the RGF package won't work properly),


and then,


to validate that a user is in the python version where RGF or FastRGF are installed. Then,


to load the R package. It is possible that the following warning in the R session appears if FastRGF is not installed,

UserWarning: Cannot find FastRGF executable files. FastRGF estimators will be unavailable for usage.
  warnings.warn("Cannot find FastRGF executable files. FastRGF estimators will be unavailable for usage.")

Windows OS [ installed / tested on Python >= 3.4 ]


First download of for windows

Update the Environment variables ( Control Panel >> System and Security >> System >> Advanced system settings >> Environment variables >> System variables >> Path >> Edit ) by adding ( for instance in case of python 3 ),


Install the Build Tools for Visual Studio

Open the Command prompt (console) and install the rgf_python dependencies,

pip3 install --upgrade setuptools
pip3 install -U numpy
pip3 install --upgrade scipy
pip3 install -U scikit-learn

Then download git for windows,

and run the downloaded .exe file. Then do,

git clone

FastRGF requires a gcc version > 5.0 . To find out the gcc version, open a command prompt (console) and type,

gcc --version

Installation / Upgrade of MinGW

Perform the following steps to upgrade the MinGW (so that simple RGF functions work – but not FastRGF)

Normally MinGW is installed in the C:\ directory. So, first delete the folder C:\MinGW (if it already exists), and then remove the environment variable from (Control Panel >> System and Security >> System >> Advanced system settings >> Environment variables >> System variables >> Path >> Edit) which usually is C:\MinGW\bin. Then download the most recent version of MinGW, and especially the mingw-get-setup.exe which is an automated GUI installer assistant. After the new version is installed successfully, update the environment variable by adding C:\MinGW\bin in (Control Panel >> System and Security >> System >> Advanced system settings >> Environment variables >> System variables >> Path >> Edit). Then open a new command prompt (console) and type,

gcc --version

to find out if the new version of MinGW is installed properly.

A word of caution, If Rtools is already installed then make sure that it does not point to an older version of gcc. Just observe the Path field of the environment variables (accessible as explained previously).

Perform the following steps only in case that a FastRGF installation is desired and gcc version is < 5.0

FastRGF works only with MinGW-w64 because only this version provides POSIX threads. It can be downloaded from the project's official SourceForge page. After a successful download and installation the user should also update the environment variables field in (Control Panel >> System and Security >> System >> Advanced system settings >> Environment variables >> System variables >> Path >> Edit) by adding the following path (assuming the software is installed in C:\Program Files (x86) folder),

C:\Program Files (x86)\mingw-w64\i686-7.2.0-posix-dwarf-rt_v5-rev1\mingw32\bin

Installation of cmake

First download cmake for Windows, win64-x64 Installer. Once the file is downloaded run the .exe file and during installation make sure to add CMake to the system PATH for all users.

Before the installation of rgf I might have to remove Rtools environment variables, such as C:\Rtools\bin (accessible as explained previously), otherwise errors might occur.

Installation of RGF, FastRGF and rgf_python [ assuming the installation takes place in the c:/ directory ]

Open a console with administrator privileges (right click on cmd and run as administrator), then

# download the most recent version of rgf-python from the GitHub repository 
git clone
# then navigate to 
cd /rgf/RGF/
mkdir bin
cd c:/
# then download the latest "rgf.exe" from and place the "rgf.exe" inside the previously created "bin" folder ( /rgf/RGF/bin )
# installation of RGF
cd /rgf/RGF/build
cd c:/
# installation of FastRGF
cd /rgf/FastRGF/
mkdir build
cd build
cmake .. -G "MinGW Makefiles"
mingw32-make install
cd c:/
# installation of rgf-python
cd rgf/python-package
python install

Then open a command prompt (console) and type,


to launch Python and then type

import rgf

to observe if rgf is installed properly. Then continue with the installation of the RGF package,


On windows the user can take advantage of the RGF package currently only from within the command prompt (console). First, find the full path of the installation location of R (possible if someone right-clicks in the R short-cut (probably on Desktop) and navigates to properties >> shortcut >> target). In case, for instance, that R is located in C:\Program Files\R\R-3.4.0\bin\x64\R, then, by opening a command prompt (console) and giving,

cd C:\Program Files\R\R-3.4.0\bin\x64\

one can proceed with the usage of the RGF package.

Installation of the RGF package

To install the package from CRAN use,


and to download the latest version from GitHub use the install_github function of the devtools package,

devtools::install_github(repo = 'RGF-team/rgf', subdir = 'R-package')

Use the following link to report bugs/issues,


RGF 1.0.6

  • We've added the init_model parameter to the RGFRegressor and RGFClassifier
  • We've added the save_model method to the RGFRegressor and RGFClassifier
  • Source files were broken up into one file per exported object as of #266
  • Internal calls to estimator constructors were changed to use keyword, instead of positional, arguments. #267

RGF 1.0.5

The RGF R package was integrated in the home repository of the Regularized Greedy Forest (RGF) library (

  • We downgraded the minimum required version of R to 3.2.0
  • We modified / formatted the R files

RGF 1.0.4

  • We modified the license from GPL-3 to MIT to go in accordance with the new structure of the rgf_python package. The package includes two files : LICENSE (for the RGF R package) and LICENSE.note (for the RGF, FastRGF and rgf_python packages).
  • We added the following new features of RGF estimators : feature_importances_ and dump_model()
  • We modified the file and especially the installation instructions for all operating systems (Linux, Mac OS X, Windows)
  • We created an R6 class (Internal_class) for all secondary functions which are used in RGF and FastRGF

RGF 1.0.3

  • The dgCMatrix_2scipy_sparse function was renamed to TO_scipy_sparse and now accepts either a dgCMatrix or a dgRMatrix as input. The appropriate format for the RGF package in case of sparse matrices is the dgCMatrix format (scipy.sparse.csc_matrix)
  • We added an onload.R file to inform the users about the previous change
  • Due to the previous changes we modified the Vignette and the tests too

RGF 1.0.2

We commented the example(s) and test(s) related to the dgCMatrix_2scipy_sparse function [ if (["sysname"] != 'Darwin') ], because the scipy-sparse library on CRAN is not upgraded and the older version includes a bug (TypeError : could not interpret data type). This leads to an error on Mac OS X ( reference : ).

RGF 1.0.1

We added links to the GitHub repository (master repository, issues).

RGF 1.0.0

Initial version.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


1.0.6 by Lampros Mouselimis, 2 months ago

Report a bug at

Browse source code at

Authors: Lampros Mouselimis [aut, cre] , Ryosuke Fukatani [cph] (Author of the python wrapper of the 'Regularized Greedy Forest' machine learning algorithm) , Nikita Titov [cph] (Author of the python wrapper of the 'Regularized Greedy Forest' machine learning algorithm) , Tong Zhang [cph] (Author of the 'Regularized Greedy Forest' and of the Multi-core implementation of Regularized Greedy Forest machine learning algorithm) , Rie Johnson [cph] (Author of the 'Regularized Greedy Forest' machine learning algorithm)

Documentation:   PDF Manual  

Task views: Machine Learning & Statistical Learning

MIT + file LICENSE license

Imports reticulate, R6, Matrix

Suggests testthat, covr, knitr, rmarkdown

System requirements: Python (2.7 or >= 3.4), rgf_python, scikit-learn (>= 0.18.0), scipy, numpy. Detailed installation instructions for each operating system can be found in the README file.

See at CRAN