README.md

Flow-based Conformal Prediction for Multi-dimensional Time Series

This repository contains soruce code of the method in the paper "Flow-based Conformal Prediction for Multi-dimensional Time Series".

Implementation Example

The code was written in Python 3.9.13 with torch 2.2.2+cu118. Additional dependencies are listed in requirements.txt.

We also provide Colab notebooks that walk through how to obtain the base predictor results and run experiments with FCP.

You can adapt these implementation examples to suit your own experiments.

Obtaining results of base predictor

In this colab notebook, we train base predictors and obtain results on wind data. GPU is not strictly required.

Implementing FCP

In this colab notebook, we can reproduce experiments using FCP on wind 2d data. GPU accelerator for colab is recommended.

Citation

@article{lee2025flow,
  title={Flow-based Conformal Prediction for Multi-dimensional Time Series},
  author={Lee, Junghwan and Xu, Chen and Xie, Yao},
  journal={arXiv preprint arXiv:2502.05709},
  year={2025}
}