SPACETIME: Causal Discovery from Non-Stationary Time Series
Authors: Sarah Mameche, Lénaïg Cornanguer, Urmi Ninad, Jilles Vreeken
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate our approach on synthetic data and real-world datasets in hydrology and meteorology. We show our results on synthetic data in Fig. 2. We report F1 scores over L (top) for changepoint detection, F1 scores over directed edges in Gτ (middle) and the DAG G (bottom) for causal discovery, and the Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI) for regime partitioning. |
| Researcher Affiliation | Academia | 1CISPA Helmholtz Center for Information Security 2Technische Universit at Berlin, Germany 3German Aerospace Center, Institute of Data Science EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 SPACETIME ... Pseudocodes of the components are deferred to the Appendix. |
| Open Source Code | Yes | 1Code available at https://eda.group/spacetime |
| Open Datasets | Yes | river discharge over different gauged catchments across Europe, given data derived from the Global Runoff Data Centre (GRDC) datasets (Cornes et al. 2018). Secondly, we study biosphere atmosphere fluxes in the FLUXNET dataset (Baldocchi 2014). |
| Dataset Splits | No | The paper describes how synthetic data is generated and the characteristics of real-world datasets, but it does not specify any training/testing/validation splits for experiments. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments. |
| Software Dependencies | No | The paper mentions "using TIGRAMITE Runge (2020)" but does not specify its version number or any other software dependencies with version numbers. |
| Experiment Setup | Yes | We sample regime changepoints L uniformly at random using a pre-set minimal duration, set to dmin 30 unless otherwise specified. We provide the datasets D from all locations to our method, using fixed hyperparameters dmin 30 and τ 2 for consistency. |