ThunderSVM ParametersΒΆ

This page is for parameter specification in ThunderSVM. The parameters used in ThunderSVM are identical to LibSVM (except some newly introduced parameters), so existing LibSVM users can easily get used to ThunderSVM.

The command line options for ThunderSVM are shown below.

  • -s: set the type of SVMs (default=0)
    • 0 – C-SVC
    • 1 – \( \nu \)-SVC
    • 2 – one-class SVMs
    • 3 – \( \epsilon \)-SVR
    • 4 – \( \nu \)-SVR
  • -t: set the type of kernel function (default=2)
    • 0 – linear: \( \boldsymbol{x}_i^T \cdot \boldsymbol{x}_j \)
    • 1 – polynomial: \( (\gamma \boldsymbol{x}_i^T \cdot \boldsymbol{x}_j + r)^d \)
    • 2 – radial basis function (RBF): \( \exp(-\gamma ||\boldsymbol{x}_i - \boldsymbol{x}_j||^2) \)
    • 3 – sigmoid: \( tanh(\gamma \boldsymbol{x}_i^T \cdot \boldsymbol{x}_j+ r) \)
  • -d: set the degree in kernel function (default=3)
  • -g: set \( \gamma \) in kernel function (default=\( \frac{1}{\text{num_features}} \))
  • -r: set \( r \) in kernel function (default=0)
  • -c: set the parameter C of C-SVC, \( \epsilon \)-SVR, and \( \nu \)-SVR (default=1)
  • -n: set the parameter \( \nu \) of \( \nu \)-SVC, one-class SVM, and \( \nu \)-SVR (\( \nu \in \)(0, 1), default=0.5)
  • -p: set the \( \epsilon \) in loss function of \( \epsilon \)-SVR (default=0.1)
  • -q: enable quiet mode
  • -m memory size: constrain the maximum memory size (MB) that thundersvm uses (default=8192)
  • -e: set tolerance of termination criterion (default=0.001)
  • -h: whether to use the shrinking heuristics (0 or 1, default=1)
  • -b: whether to train probabilistic SVC or SVR (0 or 1, default=0)
  • -wi: for weighted C-SVC, set the parameter C of class i to \( wi \times C \) (default=1)
  • -v: n-fold cross validation mode
  • -u: specify which gpu to use (default=0)
  • -o: set the number of cpu cores to use, -1 for maximum (default=-1)

The options in italic are not applicable for GPUs, and the alternative optimizations are implemented with automatically setting working set size.