Causal Discovery

CausalBench is a comprehensive benchmark suite for evaluating network inference methods on large-scale perturbational single-cell gene expression data. CausalBench introduces several biologically meaningful performance metrics and operates on two large, curated and openly available benchmark data sets for evaluating methods on the inference of gene regulatory networks from single-cell data generated under perturbations. The datasets consists of over 200000 training samples under interventions.

We provide the BCOPA-CE test set, which has balanced token distribution in the correct and wrong alternatives and increases the difficulty of being aware of cause and effect. ### construction 1. for each premise of the 500 samples in COPA-test set, we generate one event manually which is a plausible answer to the opposite question type of the original sample. 2. obtain 500 triplets of <*premise*, *cause*, *effect*> 3. construct 1000 samples by giving two different questions (cause or effect) to each triplet.

The raw data are obtained from an industrial plant for ultra-processed food production. The sampling was carried out every 5 minutes while the total production cycle takes approximately 3 hours, from raw ingredients to final semi- finished products. The extracted data represent approximately 80 days of production. Variables 2 − 14 belonging to 4 specific phases of the process and influence the qualitative variable 17. Variables 15 and 16 are external variables not controlled by the process which affect the final product. It should also be noted that some variation may be due to changes in raw materials, in production flow (variable 1) or to possible reconfiguration between weeks. However while the magnitude of effects may change between weeks, the causal relationships are dictated by the plant and process dynamics and are consistent (at the best of potential un-cofounder and faults) throughout the production .

CausalChaos! is a dataset for causal video question answering. It is based on Tom and Jerry cartoons. It features longer causal chains embedded in dynamic visual scenes. It also features challenging incorrect options, especially, Causal Confusion set which contains causally confounding incorrect options. All these factors prove to be challenging for current VLMs and other traditional Video Question Answering models.

TimeGraph is a comprehensive suite of synthetic datasets designed to benchmark causal discovery algorithms on time-series data. The dataset captures real-world complexities by incorporating temporal dynamics such as trends, seasonality, and nonstationarity, as well as sampling challenges including irregular time intervals and structured missingness. It features diverse noise types, including Gaussian, heavy-tailed, and heteroskedastic variations, and supports scenarios with latent confounding to enable evaluation under partially observed systems. The underlying causal structures span both linear and nonlinear relationships, including polynomial and trigonometric forms. The motivation behind TimeGraph is to address the current lack of robust and realistic benchmarks in time-series causal discovery. Existing datasets often overlook the intricate challenges observed in real-world domains such as Earth system science, healthcare, and economics. TimeGraph serves as a unified testbed for comparing linear and nonlinear causal discovery methods (e.g., PCMCI+, Granger causality, NOTEARS, etc), evaluating algorithmic robustness under conditions of missing data and irregular sampling, training and validating deep causal models and representation learning frameworks, and analyzing the sensitivity of methods to confounding and autocorrelation.