I’m excited to announce the release of my new Python library, CausalPlayground, developed to address critical needs in causal research. This library is the result of ongoing work that was recently accepted at the Causal and Object-Centric Representations for Robotics workshop at CVPR 2024. It provides researchers with a versatile platform to generate, sample, and interact with structural causal models (SCMs), a foundational tool in causality research.

CausalPlayground tackles a core challenge in causal research: the scarcity of real-world datasets with known causal relations. Given that much of the progress in fields like causal discovery and causal inference relies on synthetic data, the library fills a significant gap by offering fine-grained control over causal models. Researchers can create datasets of SCMs tailored to their specific needs, making it easier to train and evaluate models, all while ensuring consistency and comparability across studies.

Key features of CausalPlayground include:

• Interventional Data Generation: Easily implement experiments by intervening on causal models, crucial for tasks such as estimating causal effects.
• Interactive SCMs: Through integration with the popular Gymnasium framework, users can interact with causal models in real-time, making the library ideal for online learning and reinforcement learning applications.
• Comprehensive SCM Control: Customize every aspect of the causal model, from defining functional relationships to specifying the distributions of exogenous variables, which sets this library apart from other existing tools.
• SCM Generation at Scale: Automatically generate sets of SCMs, enabling robust quantitative research across different models and scenarios.

This project aims to accelerate research by standardizing the processes of generating and sharing synthetic data, fostering more efficient collaboration across the causal research community. The CausalPlayground library is fully open-source and available on GitHub here. It comes with detailed API documentation and welcomes contributions from the research community.

With the introduction of CausalPlayground, we hope to contribute to more rigorous and comparable studies in causality, helping researchers explore new frontiers in robotics, AI, and beyond. Keep an eye out for future updates, where I’ll be integrating even more causal models and optimizing the library for GPU-based deployments!

Feel free to explore the code and documentation, and join us in shaping the future of causal data generation!