Relational Causal Discovery

Relational Causal Discovery (RCD) is a sound and complete algorithm for learning causal models from relational data. RCD employs a novel rule, called relational bivariate orientation, that can can detect the orientation of a bivariate dependency with no assumptions on the underlying distribution. Combined with relational extensions to the rules utilized by the PC algorithm[1], RCD is provably sound and complete under the causal sufficiency assumption. Given a database and schema, RCD outputs a partially directed model that represents the equivalence class of statistically indistinguishable relational causal models.

Download RCD 1.0 distribution

RCD is designed and implemented by the Knowledge Discovery Laboratory in the College of Information and Computer Sciences at the University of Massachusetts Amherst. See “A Sound and Complete Algorithm for Learning Causal Models from Relational Data” (Maier, et al., UAI 2013) for additional information on RCD.

Knowledge Discovery Laboratory   College of Information and Computer Sciences   University of Massachusetts Amherst   Site Policies   Site Contact