cameronphchen / srm_tutorial Goto Github PK

I just find this at the tutorial page of BrainIAK for SRM:

Ideally, the number of features should be chosen using cross-validation. We will follow the lead of Chen et al. (2015) for now and use 50 features. Typically, 20 iterations should be enough for SRM to converge.

How could I choose the features by cross-validation?
How could I know which number of features is the best?

Sorry for asking such a basic question.

Thanks,
Zhao