microsoft/FLAML

Make predict() return "Undetected" in the classification estimator if confidence is below threshold

Open

#355 aperta il 23 dic 2021

Vedi su GitHub
 (0 commenti) (0 reazioni) (0 assegnatari)Jupyter Notebook (560 fork)github user discovery
enhancementhelp wanted

Metriche repository

Star
 (4364 star)
Metriche merge PR
 (Metriche PR in attesa)

Descrizione

Use case: flower recognition Input: flower photo Output: flower name

Right now if you input a flower that is not among the classes, predict() will return the one closest to it which is wrong. The right would be telling the user that it was not recognized. One way to do it is making the estimator produce a confidence float [0-1]. Then make predict() accept a confidence threshold along with the data, which if not surpassed returns "Unrecognized".

Guida contributor