fix: choose lighter dtype for distance and error stored values

oracle / macest

Model Agnostic Confidence Estimator (MACEST) - A Python library for calibrating Machine Learning models' confidence scores

Universal Permissive License v1.0

100 stars 20 forks source link

fix: choose lighter dtype for distance and error stored values #3

Open FlorentRamb opened 2 years ago

FlorentRamb commented 2 years ago

This PR tries to solve the following problem: When traning macest on a large dataset, the step of pre-computing distances and errors requires a lot of memory. A simple solution would be to use lighter dtypes: float32 for distances and bool for errors

Signed-off-by: Florent Rambaud flo.rambaud@gmail.com