Skip to content

YoungseogChung/calibrated-quantile-uq

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

44 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Quantile Methods for Calibrated Uncertainty Quantification

This is the repo for the paper Beyond Pinball Loss: Quantile Methods for Calibrated Uncertainty Quantification.

The core algorithm aims to learn the conditional quantiles of the data distribution by optimizing directly for calibration, sharpness, and adversarial group calibration.

Basic Usage

python main.py --loss scaled_batch_cal

There are currently 4 losses implemented: "calibration loss", "scaled calibration loss", "interval score", and "qr", which is simply the pinball loss. The versions of the losses with the prefix "batch" means the code for calculating the loss is
vectorized.

Bibliography

If you find this work useful, please consider citing the corresponding paper:

@article{chung2020beyond,
  title={Beyond Pinball Loss: Quantile Methods for Calibrated Uncertainty Quantification},
  author={Chung, Youngseog and Neiswanger, Willie and Char, Ian and Schneider, Jeff},
  journal={arXiv preprint arXiv:2011.09588},
  year={2020}
}

Related Repo

This work led to the development of a general uncertainty quantification repo, called Uncertainty-Toolbox. This toolbox evaluates a quantification of uncertainty based on a suite of metrics, including calibration, sharpness, adversarial group calibration, proper scoring rules, and accuracy.

About

Training quantile models

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published