CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series
Authors: Gideon Stein, Maha Shadaydeh, Jan Blunk, Niklas Penzel, Joachim Denzler
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To demonstrate the utility of Causal Rivers, we evaluate several causal discovery approaches through a set of experiments to identify areas for improvement. |
| Researcher Affiliation | Academia | Gideon Stein, Maha Shadaydeh, Jan Blunk, Niklas Penzel, Joachim Denzler Computer Vision Group Jena Friedrich Schiller University Jena Jena, Thuringia 07743, Germany EMAIL |
| Pseudocode | No | The paper describes the baseline strategies (CC, RP, Combo) in prose, detailing the logic and conditions, but does not provide structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | To make usage as accessible as possible, we provide a ready-to-use benchmark package with many features2. 2https://github.com/causalrivers |
| Open Datasets | Yes | To bridge this gap, we introduce Causal Rivers1, the largest in-the-wild causal discovery benchmarking kit for time-series data to date. Causal Rivers features an extensive dataset on river discharge... 1https://causalrivers.github.io |
| Dataset Splits | Yes | Instead, we provide sampling strategies to generate thousands of subgraphs with a flexible amount of nodes and unique graph characteristics such as single-sink nodes, root causes, hidden-confounding, or simply connected graphs. ... We used Rivers Bavaria and sampled training and validation examples (identical to the strategy Random-5 ) to finetune a pre-trained network provided by Stein et al. (2024). |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU or CPU models. |
| Software Dependencies | No | The paper mentions several methods (e.g., PCMCI, Varlingam, Dynotears, CDMI, CP) but does not provide specific version numbers for any software libraries, packages, or programming languages used. |
| Experiment Setup | Yes | As causal discovery methods typically come with at least some Hyperparameters, we performed a rudimentary Hyperparameter search per method which we document in appendix A.2. ... We performed a small Hyperparameter search, testing for different values of the learning rate, weight decay, batch size, time-series resolution, normalization, and the CP architecture. |