Precision/recall

11/15/20

---

(wrote this for similar reasons to redox)

Lets say for some class C we want to measure, we reported P positives

Precision

How many of the P’s reported are actually C?

Math: (P that are indeed C) / (P)

You can have 100% precision by reporting only 1 thing correctly as C, but missing the rest, (thus low recall).

Recall

How many of C did you actually report as P?

Math: (P that are indeed C)/ (all C)

You can have 100% recall by reporting P for everything, but precision is terrible. Sort of like when a person says a bunch of stuff that is irrelevant = “imprecise”