MLOPS-BASICS5 MIN READ
Push Back on Training-Serving Skew
Coach a teammate toward shared feature logic and parity tests.
Noah N Backend engineer 1 Name risk 2 Test 3 Fix Noah rewrote a notebook transform in service code. The visible code is simple, but the model contract may have changed. Same raw input, same feature value Parity is an invariant, not a preference. Offline validation only applies when serving recreates the trained feature meaning. Noah: It is just min-max scaling. Rewriting it in TypeScript is fine. How do you respond? Noah: What test would convince you? Pick the strongest test.
Read the full lesson
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in