Estimation Algorithms

[cthoyt]

Ananke has several classes of estimation algorithms, depending on how "challenging" the graph is (different subclasses of acyclic directed mixed graphs (ADMGs))

Step 1 would be to wrap these, step 2 is to re-implement (if necessary)

• A-fixable and MB-shielded:
  • Inverse probability weighting (IPW)
  • Outcome regression
  • Generalized AIPW
  • Efficient Generalized AIPW
• P-fixable and MB-shielded:
  1. Primal IPW (p-ipw)
  2. Dual IPW (d-ipw)
  3. APIPW (apipw)
  4. Efficient APIPW (eff-apipw)

[cthoyt]

CC @srtaheri @pnavada

[djinnome]

For estimating causal inferences, we might also want to consider tractable circuit approaches.

Wang, B., & Kwiatkowska, M. (11.04.2023). Compositional Probabilistic and Causal Inference using Tractable Circuit Models. Proceedings of The 26th International Conference on Artificial Intelligence and Statistics. AISTATS.,

Causal Discussion Group video seminar