What's new

• We have recently released ThunderGBM, a fast GBDT and Random Forest library on GPUs.
• add scikit-learn interface, see here

Overview

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, Matlab and Ruby 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.

Documentation | Installation | API Reference (doxygen)

Contents

• Getting Started
• Working without GPUs

Getting Started

Prerequisites

• 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.

• CUDA 7.5 or above

Quick Install

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

• For Linux

  • pip install thundersvm for CUDA 9.0 - linux_x86_64

  • CPU - linux_x86_64

• For Windows (64bit)

  • CUDA 10.0 - win64

  • CPU - win64

Install the Python wheel file.

pip install thundersvm-cu90-0.2.0-py3-none-linux_x86_64.whl

Example

from thundersvm import SVC
clf = SVC()
clf.fit(x, y)

Download

git clone https://github.com/Xtra-Computing/thundersvm.git

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:

cmake -DCMAKE_C_COMPILER=gcc-6 -DCMAKE_CXX_COMPILER=g++-6 ..

ThunderSVM on CPUs

# in thundersvm root directory
git submodule init eigen && git submodule update
mkdir build && cd build && cmake -DUSE_CUDA=OFF .. && 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).

@article{wenthundersvm18,
 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},
 volume={19},
 pages={797--801},
 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

• LibSVM | SVM^light | OHD-SVM | NVIDIA Machine Learning | ThunderGBM

Acknowledgement

• 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)
• Sampling Bias in Deep Active Classification: An Empirical Study. (in arXiv preprint arXiv:1909.09389, 2019).
• Machine Learning-Based Fast Banknote Serial Number Recognition Using Knowledge Distillation and Bayesian Optimization. (published in Sensors 19.19:4218, 2019).
• Texture Selection for Automatic Music Genre Classification. (in arXiv preprint arXiv:1905.11959, 2019).
• Classification for Device-free Localization based on Deep Neural Networks. (in Diss. The University of Aizu, 2019).
• An accurate and robust approach of device-free localization with convolutional autoencoder. (published in IEEE Internet of Things Journal 6.3:5825-5840, 2019).
• Accounting for part pose estimation uncertainties during trajectory generation for part pick-up using mobile manipulators. (published in IEEE International Conference on Robotics and Automation (ICRA), 2019).
• 3D semantic segmentation for high-resolution aerial survey derived point clouds using deep learning. (published in ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, 2018).
• Performance Comparison of Machine Learning Models for DDoS Attacks Detection. (published in IEEE International Computer Science and Engineering Conference (ICSEC), 2018).
• Evolving Switch Architecture toward Accommodating In-Network Intelligence. (published in IEEE Communications Magazine 58.1: 33-39, 2020).
• Kernel machines that adapt to GPUs for effective large batch training. (in arXiv preprint arXiv:1806.06144, 2018).
• Genetic improvement of GPU code. (published in IEEE/ACM International Workshop on Genetic Improvement (GI), 2019). The source code of ThunderSVM is used as a benchmark.
• Dynamic Multi-Resolution Data Storage. (published in IEEE/ACM International Symposium on Microarchitecture, 2019). The source code of ThunderSVM is used as a benchmark.