Discovering Latent Causal Graphs from Spatiotemporal Data
Authors: Kun Wang, Sumanth Varambally, Duncan Watson-Parris, Yian Ma, Rose Yu
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, SPACY outperforms state-of-the-art baselines on synthetic data, even in challenging settings where existing methods struggle, while remaining scalable for large grids. SPACY also identifies key known phenomena from real-world climate data. |
| Researcher Affiliation | Academia | 1Department of Computer Science and Engineering, University of California, San Diego, La Jolla, USA 2Halıcıoglu Data Science Institute, University of California, San Diego, La Jolla, USA 3Scripps Institution of Oceanography, University of California, San Diego, La Jolla, USA. |
| Pseudocode | No | The paper includes mathematical formulations, generative models, and probabilistic graphical models (Figure 3, Figure 7), but it does not contain any explicitly labeled pseudocode or algorithm blocks with structured steps. |
| Open Source Code | Yes | An implementation of SPACY is available at https: //github.com/Rose-STL-Lab/SPACY/ |
| Open Datasets | Yes | The Global Climate Dataset (Baker et al., 2019) is a mixed real-simulated dataset containly monthly global temperature and precipitation from 1999 to 2001. |
| Dataset Splits | Yes | We use an 80/20 training and validation split to evaluate the validation likelihood during training. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments, such as GPU/CPU models, memory, or cloud instances. It mentions computational costs in Figure 5 but no hardware specifications. |
| Software Dependencies | No | The paper refers to various methods and frameworks by their paper citations (e.g., Rhino (Gong et al., 2023), Gumbel-Softmax trick (Jang et al., 2017), rational spline flow model (Durkan et al., 2019)), but it does not specify any software libraries or packages with version numbers that would be required to reproduce the experiments. |
| Experiment Setup | Yes | Table 2: Table showing the hyperparameters used with SPACY. For the Synthetic, Synthetic-Multivariate, and Global Climate datasets, the outer augmented Lagrangian (auglag) steps are set to 60, with a maximum of 6000 inner auglag steps. For Synthetic datasets we used batch size of 100 samples per training, and 500 for Global Climate datasets. |