Tunning parammeter C of SVM OCAS solver
In this experiment it will be used Support Vector Machine in its Extensive form  for learning linear model. So the aim of this experiment will be to learn model (1) and tune its parameter C. This parameter specify how much will classifier be fitted to trainning samples. If C is too low then classifier is more general and reversely, if C is too much high then classifier is overfitted.
Equation (1): Extensive Support Vector Machine model
For learning this model it will be used free redistributable library libOCAS  that is based on cutting planes.
For measuring error we using protocol based on three variuos dataset. The first dataset is the biggest (about 12 thousand samples) and classifier is trained on it for each parameter (Traning dataset). On next dataset (Validation dataset) we measure error on trained classifiers and then choose that, which validation error is the lowest. In the last step, we test classifier selected in previous step on the last set named Testing dataset. Each dataset should be with high diversity. This protokol will be used in all further experiments.
|Parameters of experiment|
|imSize [h x w]||80 x 64|
|winSize [h x w]||60 x 40|
Resources Corinna Cortes, Vladimir Vapnik: Support-vector networks - http://www.springerlink.com/content/k238jx04hm87j80g/
 Vojtech Franc, Soeren Sonnenburg: LIBOCAS - Library implementing OCAS solver for training linear SVM classifiers from large-scale data - http://cmp.felk.cvut.cz/~xfrancv/ocas/html/