You're training a supervised model and have a large set of candidate features, some of which may be redundant, noisy, or unstable across samples.
How would you approach feature selection for a model?
