Skip to content

Latest commit

 

History

History
37 lines (29 loc) · 945 Bytes

README.md

File metadata and controls

37 lines (29 loc) · 945 Bytes

TensorSVM

TensorSVM is a high performance GPU based SVM tool.

How to train

Sample cmdline:

Disabling TensorCore

./tensorsvm-train -k 32 -c 32 -g 0.01 -t 2 SUSY.training SUSY.model

Enabling TensorCore

./tensorsvm-train -tensorcore -k 32 -c 32 -g 0.01 -t 2 SUSY.training SUSY.model

Parameters

Rank: -k C: -c gamma: -g type: -t 0 Linear kernel, -t 2 RBF kernel Enabling TensorCore: -tensorcore 'SUSY.training' is training file path 'SUSY.model' is output model

How to predict

We are using ThunderSVM https://github.com/Xtra-Computing/thundersvm to predict.

./thundersvm-predict SUSY.testing SUSY.model output

Sample cmdline:

./tensorsvm-train-mixed -k 400 -c 32 -g 0.0002 -t 2  /home/pwu6/LIBSVM_DATA/covtype.libsvm.binary.1 && /home/pwu6/thundersvm/build/bin/thundersvm-predict /home/pwu6/LIBSVM_DATA/covtype.libsvm.binary.2  /home/pwu6/LIBSVM_DATA/covtype.libsvm.binary.1.model output