Interventional Characteristic Imsets

In the area of causal discovery, one seeks to use data to learn the directed acyclic graph (DAG) that best explains the causal relationship among random variables. One method of learning this graph is via a linear program over the characteristic imset (CIM) polytope. We introduce an interventional analogue of the CIM polytope and show that interventional data can be used to learn the DAG via a linear program. We also use this framework to compute an H-representation for faces of CIM polytopes given by restricting graphs to have a fixed tree skeleton.