Estimand-Agnostic Causal Query Estimation with Deep Causal Graphs

Causal Queries are usually estimated by means of an estimand, a formula consisting observational terms that can be computed using passive data. Each query results in different formula, which makes estimand-based methods extremely ad-hoc. In this work, we propose estimand-agnostic framework capable computing any identifiable causal on arbitrary Graph (even the presence latent confounders) with only one general model. We provide multiple implementations leverage expressive power Neural Networks and Normalizing Flows to model complex distributions, derive estimation procedures for all kinds observational, interventional counterfactual queries, valid kind graph is identifiable. Finally, test our techniques modelling setting benchmark show how, despite being query-agnostic framework, it compete query-specific models. Our proposal includes open-source library allows easy application extension researchers practitioners alike.