https://github.com/msesia/conditional-conformal-pvalues
Conditional calibration of conformal p-values for outlier detection.
https://github.com/msesia/conditional-conformal-pvalues
conformal-prediction false-discovery-rate machine-learning outlier-detection statistics
Last synced: 3 months ago
JSON representation
Conditional calibration of conformal p-values for outlier detection.
- Host: GitHub
- URL: https://github.com/msesia/conditional-conformal-pvalues
- Owner: msesia
- Created: 2020-12-18T17:31:49.000Z (about 5 years ago)
- Default Branch: main
- Last Pushed: 2022-11-15T07:09:08.000Z (about 3 years ago)
- Last Synced: 2023-05-10T14:18:11.728Z (over 2 years ago)
- Topics: conformal-prediction, false-discovery-rate, machine-learning, outlier-detection, statistics
- Language: Python
- Homepage:
- Size: 305 KB
- Stars: 17
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Testing for Outliers with Conformal p-values
We study the construction of p-values for nonparametric outlier detection, taking a multiple-testing perspective. The framework is that of conformal prediction, which wraps around any machine-learning algorithm to provide finite-sample guarantees regarding the validity of predictions for future testpoints. In this setting, existing methods can compute p-values that are marginally valid but mutually dependent for different future test points.
This repository contains a software implementation and guided examples for the methodology developed in the [accompanying paper](https://arxiv.org/abs/2104.08279), which provides a new method to compute p-values that are both conditionally valid and independent of each other for different future test points, thus allowing multiple testing with stronger stronger type-I error guarantees.
```
"Testing for Outliers with Conformal p-values"
Stephen Bates, Emmanuel Candes, Lihua Lei, Yaniv Romano, and Matteo Sesia.
accepted in Annals of Statistics (2022)
arXiv pre-print: https://arxiv.org/abs/2104.08279
```