Sequential Predictive Conformal Inference for Time Series

Authors: Chen Xu, Yao Xie

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Using simulation and real-data experiments, we demonstrate a significant reduction in interval width of SPCI compared to other existing methods under the desired empirical coverage.Experimentally, we demonstrate competitive and/or improved empirical performance against baseline CP methods on sequential data.
Researcher Affiliation Academia 1H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA. Correspondence to: Yao Xie <yao.xie@isye.gatech.edu>.
Pseudocode Yes Algorithm 1 Sequential Predictive Conformal Inference (SPCI)Algorithm 2 SPCI for exchangeable data (based on split conformal)Algorithm 3 Multi-step SPCI (based on LOO prediction in Enb PI (Xu & Xie, 2021b))
Open Source Code Yes Official implementation can be found at https://github.com/hamrel-cxu/SPCI-code.
Open Datasets Yes The first dataset is the wind speed data (m/s) at wind farms operated by the Midcontinent Independent System Operator (MISO) in the US (Zhu et al., 2021). The second dataset contains solar radiation information1 in Atlanta downtown, which is measured in Diffuse Horizontal Irradiance (DHI). The full dataset contains a yearly record in 2018 and is updated every 30 minutes. 1Collected from National Solar Radiation Database (NSRDB): https://nsrdb.nrel.gov/. The last dataset tracks electricity usage and pricing (Harries et al., 1999) in the states of New South Wales and Victoria in Australia, with an update frequency of 30 minutes over a 2.5-year period in 1996 1999.Specifically, the dataset is publicly available on Kaggle https://www.kaggle.com/datasets/paultimothymooney/stock-market-data
Dataset Splits Yes We fix α = 0.1 and use the first 80% (resp. rest 20%) data for training (resp. testing).
Hardware Specification No The paper does not specify any hardware details such as GPU or CPU models used for the experiments.
Software Dependencies No In our experiments, we use the Python implementation of QRF by (Roebroek, 2022).Roebroek, J. Sklearn-quantile, 2022. URL https://github.com/jasperroebroek/sklearn-quantile.
Experiment Setup Yes We fix α = 0.1 and use the first 80% (resp. rest 20%) data for training (resp. testing). For SPCI and Enb PI, we use the random forest regression model with 25 bootstrap models.