Skip to content

Latest commit

 

History

10 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Conformal prediction

This repository is a playground for conformal prediction (CP), a statistical framework for uncertainty quantification in machine learning. CP relies on prediction sets or intervals of possible outcomes so as to represent the uncertainty of model predictions. These sets are constructed such that they are guaranteed to contain the true outcome with a pre-specified probability.

What makes CP attractive and broadly applicable is that it comes with strong guarantees under widely met conditions. The theoretical properties do not hinge on the amount of data (finite-sample) and virtually require no distributional assumptions (distribution-free). Moreover, any model can be used (model-agnostic).

APS predictions sets are visualized for a 2D 3-way classification problem CQR prediction intervals are shown for a 1D regression problem with heteroscedastic noise

Notebooks

Releases

Packages

Contributors

Languages