BoostSVM

Overview

BoostSVM is a tool for efficient training of SVM on GPUs, which optimizes the kernel matrix computation. It is an optimization of ThunderSVM.

Prerequisites

• cmake 2.8 or above
• gcc 4.8 or above for Linux
• CUDA 11.0+

Introduction

Download

git clone https://github.com/BoruiXu/BoostSVM.git

Build on Linux.

cd BoostSVM
mkdir build
cd build
cmake ..
make -j

Usage example

./bin/thundersvm-train -c 100 -g 0.5 dataset_path

Datasets

All test datasts can be downloaded through the script in dataset floader.

sh ./dataset/get_datasets.sh

Parameter and Test

The meaning of parameters is the same as that in ThunderSVM. We provide a bash script (train_run.sh) to train datasets mentioned in our paper. Befor running this script, please copy this script into the build floder. And we use different number standing for different datasets. For example, if you want to test the real-sim dataset. Please run:

sh train_run.sh 1

NOTED: all datasets need to be stored in the dataset folder!