Causal polytrees are singly connected causal models and they arefrequently applied in practice. However, in various applications, manyvariables remain unobserved and causal polytrees cannot be appliedwithout explicitly including unobserved variables. Our study thusproposes the ancestral polytree model, a novel combination ofancestral graphs and singly connected graphs. Ancestral graphs canmodel causal and non-causal dependencies, while singly connectedmodels allow for efficient learning and inference. We discuss thebasic properties of ancestral polytrees and propose an efficientstructure learning algorithm. Experiments on synthetic datasets andbiological datasets show that our algorithm is efficient and theapplications of ancestral polytrees are promising.
展开▼
