BICause Trees

The repo containing the code and experiments for the model described in Hierarchical Bias-Driven Stratification for Interpretable Causal Effect Estimation; accepted to AISTATS 2025.

The code in this reporsitory cotains the version used to obtain the results from the paper.

Causallib version

However, version 0.10.0 of causallib now contains a cleaner, better-tested, and improved version of the proposed model (see BiomedSciAI/causallib#76).

We therefore encourage people using the version existing from causallib if they intend to use the model for their own research.

This can be done by the following:

1. Install causallib:

   pip install causallib

2. Import BICauseTree:

   from causallib.contrib.bicause_tree import BICauseTree

3. At this point BICauseTree behaves like any other IndividualOutcomeEstimator in causallib.
   This means that, given covaraites X, treatment assignment a, and outcome y, one can

   bic = BICauseTree()  # See documentation for parametrization
   bic.fit(X, a, y)  # Fit the causal model based on the tree
   avg_outcomes = bic.estimate_population_outcome(X, a, y)
   ind_outcomes = bic.estimate_individual_outcome(X, a, y)
   # Specifying `y` may be optional depending on what type of `outcome_model` was passed to BICauseTree.