Hacker News new | past | comments | ask | show | jobs | submit login

The argument is you can design a model, check that it runs, then submit the job to train / evaluate the model on real data. The data scientist doesn't necessarily need to touch the real data ever (especially if ML is used).

This is not a great setup if one requires a model that can extrapolate.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact