Conformal prediction interval for dynamic time-series

Authors: Chen Xu, Yao Xie

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform extensive real-data analyses to demonstrate its effectiveness.Empirically, we extensively study the performance of Enb PI on the renewable energy estimation application, using solar and wind data.5. Experimental Results
Researcher Affiliation Academia 1Industrial and Systems Engineering, Georgia Institute of Technology. Correspondence to: Chen Xu <cxu310@gatech.edu>, Yao Xie <yao.xie@isye.gatech.edu>.
Pseudocode Yes Algorithm 1 Sequential Distribution-free Ensemble Batch Prediction Intervals (Enb PI)
Open Source Code No The paper does not contain any explicit statement about releasing source code or a link to a code repository.
Open Datasets No The paper mentions "2018 hourly solar radiation data from Atlanta and 9 cities in California, as well as 2019 hourly wind energy data from the Hackeberry wind farm in Austin" but does not provide specific links, DOIs, repositories, or formal citations for public access to these datasets.
Dataset Splits Yes We do a 50:50 split into proper training set and calibration set for ICP and Weighted ICP.we use the first 20% of total hourly data for training unless otherwise specified.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions "Python s statsmodel package", "Python sklearn library", and "keras library" but does not provide specific version numbers for these software components.
Experiment Setup Yes Throughout this subsection, we fix s = 1, so every observation comes in sequence without delay. We let α = 0.1 and use the first 20% of total hourly data for training unless otherwise specified. Lastly, we use Enb PI under B = 30 and φ as taking the sample mean.All parameters into Enb PI except choices of s are kept the same unless otherwise specified. We pick s = 14 for Atlanta solar dataSee Section 8.1 for their parameter specifications.