This repository contains codes for implementing models and reproducing the results in our paper "Transformer Conformal Prediction for Time Series". Some parts of the codes are inherited from "Sequential Predictive Conformal Inference for Time Series".

Transformer Conformal Prediction for Time Series
    Junghwan Lee, Chen Xu, and Yao Xie
    ICML 2024 Workshop on Structured Probabilistic Inference & Generative Modeling
    https://arxiv.org/abs/2406.05332

Sequential Predictive Conformal Inference for Time Series
    Chen Xu and Yao Xie
    ICML 2023
    https://proceedings.mlr.press/v202/xu23r/xu23r.pdf

The codes were written in Python 3.9.13. If you want to implement locally,

1. Clone this repository

2. Move to the directory where you clone this repository and install requirements using pip

    pip install -r requirements.txt
   

We also provide colab notebook for reproducing the results. Note that results from colab notebook shows slightly different performance from the results in the paper.

Conformal prediction on time series with heteroscedastic errors as described in Section 4.2 in the paper.

Conformal prediction on the electricity dataset in Section 4.3 in the paper.