Efficient Bayesian Learning Curve Extrapolation using Prior-Data Fitted Networks

This repository offers an implementation of LC-PFN, a method designed for efficient Bayesian learning curve extrapolation.

LC-PFN in action on Google colab and HuggingFace

Installation using pip:

pip install -U lcpfn

Update: there is an inconsistency between the code and the paper regarding the definition of the noise prior. The correct definition is the one used in the code, where $\log(\sigma)$ is defined as $\mathcal{N}(-4, 1)$.

Usage

Try out the notebooks (require matplotlib) for training and inference examples.

NOTE: Our model supports only increasing curves with values in $[0,1]$. If needed, please consider normalizing your curves to meet these constraints. See an example in notebooks/curve_normalization.ipynb.

Reference

@inproceedings{
adriaensens2023lcpfn,
title={Efficient Bayesian Learning Curve Extrapolation using Prior-Data Fitted Networks},
author={Adriaensen, Steven and Rakotoarison, Herilalaina and Müller, Samuel and Hutter, Frank},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=xgTV6rmH6n}
}