Encoding quick reference
Recall when to use one-hot, rare bucketing, target encoding, and unknown fallbacks.
Common objection The ID hash performs best, so the model must be learning useful merchant behavior. A high-cardinality identifier is dominating validation. Your line It may be memorizing historical entities. Let us test support, new-merchant performance, and a smoothed encoding before trusting the lift. Treating identifier memorization as stable behavior signal. It keeps the signal candidate but demands generalization evidence. Encoding choice When should you prefer each approach? The right encoding states fit boundary, support threshold, and serving fallback. Unseen values What should happen to a category first seen in production? Route it to a defined unknown bucket or learned…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in