looms: leave-one-out model selection for support vector machines

Jen-Hao Lee and Chih-Jen Lin

Please note that looms supports only model selection for two-class problems. Now we have provided a simpler model selection tool (for multi-class svm) in libsvm. Please use it instead.

Automatic model selection is an important issue to make support vector machines (SVM) practically useful. Most existing approaches use the leave-one-out (loo) related estimators which are considered computationally expensive. looms uses some numerical tricks which lead to efficient calculation of loo rates of different models.

Given a range of parameters, looms automatically returns the parameter and model with the best loo rate. For example,

% looms heart_scale
Optimal parameter: c=16.000000, gamma=0.016000, rate= 83.704%

where c is the penalty parameter (or say the upper bound of the SVM dual formulation) and gamma is the parameter of the RBF kernel: exp(gamma*|x_i - x_j|^2). Currently we support only the RBF kernel.

looms is based on the software BSVM by Chih-Wei Hsu and Chih-Jen Lin.
Details of looms appear in the following paper:
J.-H. Lee and C.-J. Lin, Automatic model selection for support vector machines .

Download looms

Note that looms is provided "as is" without express or implied warranty. This software can be freely used for research purpose. Use for commercial purposes is expressly prohibited without contacting the authors.

An Example

Acknowledgments: This work was supported in part by the National Science Council of Taiwan via the grant NSC 89-2213-E-002-013. The authors thank Chih-Wei Hsu for many helpful discussions and comments.