Hacker News new | ask | show | jobs
by svg7 252 days ago
You just evaluate it against whatever test data you used and compute a bunch of metrics. You decide to use the model, if "bad things" happen at an acceptable enough rate.