Skip to content

Pytorch library for model calibration metrics and visualizations as well as recalibration methods. In progress!

Notifications You must be signed in to change notification settings

Jonathan-Pearce/calibration_library

Repository files navigation

Calibration Library

Jonathan Pearce, McGill University

A model calibration library currently under construction. Built for PyTorch models, this library enables users to evaluate their model's uncertainty estimates (probability estimates) using popular calibration metrics, train model wrappers that improve model calibration and generate data visualizations to identify where and how their model's are well calibrated or not.

Diagram

Diagram2

References

Metrics

ECE and MCE - Obtaining Well Calibrated Probabilities Using Bayesian Binning

SCE, ACE and TACE - Measuring Calibration in Deep Learning

Recalibration Methods

Tempurature Scaling - On Calibration of Modern Neural Networks

Visualizations

Reliability Diagram and Confidence Histograms - On Calibration of Modern Neural Networks

About

Pytorch library for model calibration metrics and visualizations as well as recalibration methods. In progress!

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages