BCNP · Bayesian Causal Discovery

Meta-learned posterior over causal DAGs in one forward pass

Use when you have observational data across several variables and want a Bayesian posterior over the causal DAG — not a single MAP graph. Especially useful when the true structure is unidentifiable (Markov equivalence class, few samples) and multiple DAGs plausibly explain the data, because BCNP returns calibrated uncertainty with edge dependencies intact.