BayOTIDE: Bayesian Online Multivariate Time Series Imputation with Functional Decomposition

Authors: Shikai Fang, Qingsong Wen, Yingtao Luo, Shandian Zhe, Liang Sun

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We extensively evaluate our method on synthetic and real-world datasets, and the results show that Bay OTIDE outperforms the state-of-the-art methods in terms of accuracy and efficiency.
Researcher Affiliation Collaboration 1University of Utah, USA 2DAMO Academy, Alibaba Group 3Carnegie Mellon University, USA.
Pseudocode Yes Algorithm 1 Bay OTIDE
Open Source Code Yes We release the code at https://github.com/ xuangu-fang/Bay OTIDE.
Open Datasets Yes Traffic-Guangzhou(Chen et al.): traffic speed records in Guangzhou with 214 channels and 500 timestamps. Solar-Power(https://www.nrel.gov/ grid/solar-power-data.html) : 137 channels and 52560 timestamps, which records the solar power generation of 137 PV plants. Uber-Move(https:// movement.uber.com/): 7489 channels and 744 timestamps, recording the average movement of Uber cars along with the road segments in London, Jan 2020.
Dataset Splits No The paper states, 'For each dataset, we randomly sample {70%, 50%} of the available data points as observations for model training, and the rest for evaluation.' It mentions training and evaluation, but does not explicitly define a separate validation set or its split.
Hardware Specification No The paper does not provide specific details about the hardware used for the experiments, such as particular GPU or CPU models, or memory specifications.
Software Dependencies No The paper mentions 'implemented it by Pytorch' but does not specify a version number for PyTorch or any other software libraries or dependencies.
Experiment Setup Yes Detailed information on hyperparameter settings is provided at Table 5 in the appendix.