|
Hi, I am trying to train a classifier performing semi-supervised learning and feature selection. I am using manifold regularization but I am stuck in the feature selection part. Manifold regularization also depends on some free parameters so I have to perform cross-validation too. 1) Is there any semi-supervised algorithms that perform feature selection and learning at the same time. I mean the output is a classifier with that uses only some subset of features. 2) Would it be correct to apply manifold regularization with backward feature selection? I mean I have n features. I choose all subsets of n-1, and I train and CV to select the best model for each n-1 subset of features. then I stay with the one that has the best accuracy and do the same for n-2. Is that correct? Thanks |
