Skip to content

Comment on Successful machine learning models: lessons learned at Booking.comparent

Comments

It's even worst than that. Most of the time the validity of the t-test that they are running is questionable. They are technically running an online t-test and as soon as they find significance they stop. This is fundamentally wrong, and not conclusive at all.

A few years ago when I was still working there, involved in the experimentation tooling among other things, we largely excised that particular behavior. What are you basing your assertion on?

Cheating on a hypothesis test... that's terrible.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.