Implementation of the IMP and IMP_inv algorithms from the work
"Learning invariant representations under general interventions on the response" by Kang Du and Yu Xiang.
Data Generation: Generate data for Experiment A-3 from [1].
training_data.m: Generate training data.
testing_data.m: Generate test data.
Algorithms:
IMP_inv_training.m: The training procedure of IMP.
IMP_training.m: The training procedure of IMP_inv.
IMP_testing.m: The testing procedure of IMP and IMP_inv.
run_example.m: Run examples of Experiment A-3.
Utility Functions:
ols.m: Compute OLS estimator.
multi_ols.m: Compute OLS for each environment.
residual_test.m: Test the invariance of the residual (procedure II from [2]).
References:
[1] Du, Kang, and Yu Xiang. "Learning invariant representations under general interventions on the response." arXiv preprint arXiv:2208.10027 (2022).
[2] Peters, Jonas, Peter Bühlmann, and Nicolai Meinshausen. "Causal inference by using invariant prediction: identification and confidence intervals." Journal of the Royal Statistical Society Series B: Statistical Methodology 78.5 (2016): 947-1012.