Conformal prediction for multi-dimensional time series by ellipsoidal sets

Authors: Chen Xu, Hanyang Jiang, Yao Xie

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we demonstrate that Multi Dim SPCI maintains valid coverage on a wide range of multivariate time series while producing smaller prediction regions than CP and non-CP baselines.
Researcher Affiliation Academia 1H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology.
Pseudocode Yes Algorithm 1 Multi-dimensional SPCI (Multi Dim SPCI)
Open Source Code Yes Code is available at https: //github.com/hamrel-cxu/Multi Dim SPCI.
Open Datasets Yes We consider three real multivariate time-series datasets. The first wind dataset considers wind speed in meters per second at different wind farms (Zhu et al., 2021) [...] The second solar dataset considers solar radiation in Diffused Horizontal Irradiance (DHI) units at different solar sensors (Zhang et al., 2021) [...] The third traffic dataset considers traffic flow collected at different traffic sensors (Xu & Xie, 2021a).
Dataset Splits Yes The initial 80K samples {Yt} are training data; the remaining 20K samples are test data. [...] The first 85% data are used for training, and the remaining 15% are used for testing.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. It only mentions 'running experiments' in general terms.
Software Dependencies No The paper does not specify version numbers for any software dependencies, libraries, or frameworks used in the experiments (e.g., Python, PyTorch, TensorFlow, etc.).
Experiment Setup Yes In all our experiments, the value of ρ used in (4) is set to be 0.001. [...] For simplicity, we only consider the global covariance matrix in (3) rather than its local variant (A.1), which would bring further improvements. [...] We recommend setting k = 0.1T and λ = 0.95 to capture local variations effectively. [...] The multivariate linear regression method is used as the point predictor.