Skip to content
ThunderSVM: A Fast SVM Library on GPUs and CPUs
C++ Python Cuda CMake R C Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
Matlab fix bug #64 May 9, 2018
R Altered the error message Aug 30, 2019
docs Update requirements.txt Jun 3, 2019
eigen @ dde02fc
include fix predict output on cmd May 27, 2019
src fix predict output on cmd May 27, 2019
.gitignore Added Python package Jun 28, 2018
.travis.yml add option for building tests Mar 18, 2018
Doxyfile update comments Nov 23, 2017
LICENSE Create LICENSE Nov 1, 2017 add whl files Oct 28, 2019
appveyor.yml enable test on windows Mar 28, 2018
thundersvm-full.pdf improve supplementary material Jun 10, 2018

Build Status Build status GitHub license Documentation Status GitHub issues PyPI version

What's new

  • We have recently released ThunderGBM, a fast GBDT and Random Forest library on GPUs.
  • add scikit-learn interface, see here
  • pre-built binaries and DLL for Windows x64 on CPUs are avaliable


The mission of ThunderSVM is to help users easily and efficiently apply SVMs to solve problems. ThunderSVM exploits GPUs and multi-core CPUs to achieve high efficiency. Key features of ThunderSVM are as follows.

  • Support all functionalities of LibSVM such as one-class SVMs, SVC, SVR and probabilistic SVMs.
  • Use same command line options as LibSVM.
  • Support Python, R and Matlab interfaces.
  • Supported Operating Systems: Linux, Windows and MacOS.

Why accelerate SVMs: A survey conducted by Kaggle in 2017 shows that 26% of the data mining and machine learning practitioners are users of SVMs.

Documentations | Installation | API Reference (doxygen)


Getting Started


  • cmake 2.8 or above
  • gcc 4.8 or above for Linux and MacOS
  • Visual C++ for Windows

If you want to use GPUs, you also need to install CUDA.

Quick Install

Download the Python wheel file (For Python3 or above).

Install the Python wheel file.

pip install thundersvm-cu90-0.2.0-py3-none-linux_x86_64.whl
from thundersvm import SVC
clf = SVC(), y)


git clone

Build on Linux (build instructions for MacOS and Windows)

ThunderSVM on GPUs
cd thundersvm
mkdir build && cd build && cmake .. && make -j

If you run into issues that can be traced back to your version of gcc, use cmake with a version flag to force gcc 6. That would look like this:

ThunderSVM on CPUs
# in thundersvm root directory
git submodule init eigen && git submodule update
mkdir build && cd build && cmake -DUSE_CUDA=OFF -DUSE_EIGEN=ON .. && make -j

If make -j doesn't work, please simply use make. The number of CPU cores to use can be specified by the -o option (e.g., -o 10), and refer to Parameters for more information.

Quick Start

./bin/thundersvm-train -c 100 -g 0.5 ../dataset/test_dataset.txt
./bin/thundersvm-predict ../dataset/test_dataset.txt test_dataset.txt.model test_dataset.predict

You will see Accuracy = 0.98 after successful running.

How to cite ThunderSVM

If you use ThunderSVM in your paper, please cite our work (full version).

 author = {Wen, Zeyi and Shi, Jiashuai and Li, Qinbin and He, Bingsheng and Chen, Jian},
 title = {{ThunderSVM}: A Fast {SVM} Library on {GPUs} and {CPUs}},
 journal = {Journal of Machine Learning Research},
 year = {2018}

Other publications

  • Zeyi Wen, Jiashuai Shi, Bingsheng He, Yawen Chen, and Jian Chen. Efficient Multi-Class Probabilistic SVMs on GPUs. IEEE Transactions on Knowledge and Data Engineering (TKDE), 2018.
  • Zeyi Wen, Bingsheng He, Kotagiri Ramamohanarao, Shengliang Lu, and Jiashuai Shi. Efficient Gradient Boosted Decision Tree Training on GPUs. The 32nd IEEE International Parallel and Distributed Processing Symposium (IPDPS), pages 234-243, 2018.

Related websites


  • We acknowledge NVIDIA for their hardware donations.
  • This project is hosted by NUS, collaborating with Prof. Jian Chen (South China University of Technology). Initial work of this project was done when Zeyi Wen worked at The University of Melbourne.
  • This work is partially supported by a MoE AcRF Tier 1 grant (T1 251RES1610) in Singapore.
  • We also thank the authors of LibSVM and OHD-SVM which inspire our algorithmic design.

Selected projects that use ThunderSVM

  • Scene Graphs for Interpretable Video Anomaly Classification (published in NeurIPS18)
  • 3D Semantic Segmentation for High-resolution Aerial Survey Derived Point Clouds using Deep Learning (published in SIGSPATIAL’18)
  • Accounting for part pose estimation uncertainties during trajectory generation for part pick-up using mobile manipulators. (published in International Conference on Robotics and Automation (ICRA), 2019).
You can’t perform that action at this time.