causalAssembly
收藏资源简介:
causalAssembly是由慕尼黑工业大学和Robert Bosch GmbH合作开发的一个半合成数据生成工具,用于评估因果发现方法。该数据集基于一个复杂的真实世界数据集,该数据集包含在制造环境中的装配线上的测量数据。通过详细研究装配线物理过程的基础,研究人员建立了一组部分真实的因果关系。为了解决潜在的混杂因素和隐私问题,使用分布随机森林来估计和表示由真实因果图暗示的条件分布。这些条件分布被组合成一个严格遵循观测变量因果模型的联合分布。通过从该分布中采样,causalAssembly生成的数据保证与真实情况下的马尔可夫性质一致。该工具展示了如何基准测试几种著名的因果发现算法,适用于复杂的生产数据,旨在解决因果关系学习中的验证问题。
causalAssembly is a semi-synthetic data generation tool jointly developed by the Technical University of Munich and Robert Bosch GmbH, designed for evaluating causal discovery methods. This tool is built upon a complex real-world dataset containing measurement data collected from assembly lines in manufacturing environments. By thoroughly investigating the underlying physical processes of the assembly line, researchers established a set of partially realistic causal relationships. To address potential confounding factors and privacy concerns, distributional random forests are employed to estimate and represent the conditional distributions implied by the real causal graph. These conditional distributions are then combined into a joint distribution that strictly adheres to the causal model of the observed variables. By sampling from this joint distribution, the data generated by causalAssembly is guaranteed to be consistent with the Markov property in real-world scenarios. This tool demonstrates how to benchmark several well-known causal discovery algorithms for complex manufacturing data, aiming to address the validation challenges in causal relationship learning.

- 1$\texttt{causalAssembly}$: Generating Realistic Production Data for Benchmarking Causal Discovery慕尼黑工业大学 · 2024年



