Kernel-based Optimally Weighted Conformal Time-Series Prediction
Authors: Jonghyeok Lee, Chen Xu, Yao Xie
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the superior performance of KOWCPI on real time-series against state-of-the-art methods, where KOWCPI achieves narrower confidence intervals without losing coverage. ... We demonstrate the effectiveness of KOWCPI on real time-series data against state-of-the-art baselines. |
| Researcher Affiliation | Academia | Jonghyeok Lee, Chen Xu & Yao Xie H. Milton Stewart School of Industrial and Systems Engineering Georgia Institute of Technology Atlanta, GA 30332, USA EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 Kernel-based Optimally Weighted Conformal Prediction Intervals (KOWCPI) |
| Open Source Code | No | The paper does not provide an explicit statement about releasing the source code for KOWCPI or a link to its repository. It only mentions code for baselines: "For the implementation of ACI-related methods, we utilized the R package Adaptive Conformal (https://github.com/herbps10/Adaptive Conformal). For SPCI and Enb PI, we used the Python code from https://github.com/hamrel-cxu/SPCI-code." |
| Open Datasets | Yes | The first ELEC2 data set (electric) (Harries, 1999) ... The second renewable energy data (solar) (Zhang et al., 2021) ... The third wind speed data (wind) (Zhu et al., 2021) ... We compare KOWCPI with baseline methods using Apple s daily closing stock price data from January 1, 2020, to December 12, 2022. This publicly available dataset can be accessed on Kaggle (https: //www.kaggle.com/datasets/paultimothymooney/stock-market-data). |
| Dataset Splits | Yes | Every dataset is split in a 7:1:2 ratio for training the point predictor, tuning the window length w and bandwidth h, and constructing prediction intervals, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions using "random forest as the base point predictor" and for baselines "R package Adaptive Conformal" and "Python code", but does not provide specific version numbers for any software dependencies directly used for KOWCPI or its implementation. |
| Experiment Setup | Yes | In all comparisons, we use the random forest as the base point predictor with the number of trees = 10. Every dataset is split in a 7:1:2 ratio for training the point predictor, tuning the window length w and bandwidth h, and constructing prediction intervals, respectively. The window length for each dataset is fixed and determined through cross-validation, while the bandwidth is selected by minimizing the nonparametric AIC, as detailed in equation 11. |