|
Bayesian model selection for SVM/SVR following Kwok et al. ,
|
Proceedings of the European Symposium on Artificial Neural Networks (ESANN), pp.177-182, Bruges, Belgium, April 1999.
for selecting the parameter C and ranking models
inputs:
A - SVM/SVR
hyperparameters and their defaults:
use_balanced_C=0 - adapt balanced C for SVM
type='L1' - use l1-SVM
Methods:
train - optimizes soft margin parameter for given data
test -
outputs:
pbest - best parameter C
posterior - posterior probability for the model (given the data)
Example:
[r,a]=train(bayessel(svm),gen(toy))
[a.pbest,a.posterior]
|
Reference : (The Evidence Framework Applied to SVM / Bayesian SVR) |