-
LobsterNet is a multitask neural network designed for heterogeneous treatment assignment effect estimation under treatment non-adherence.
-
It leverages the conditional front-door adjustment (CFD), which theoretically gaurantees lower variance estimates than the commonly used standard backdoor adjustment (CFD) when the true treatment effect is small.
-
This repository contains reproducible codes for experimental results in "Heterogeneous Treatment Assignment Effect Estimation Under Non-adherance with Conditional Front-door Adjustment" (CHIL 2025).
- Synthetic datasets will be automatically generated in the following steps
- IHDP dataset is provided under
data/IHDP/ihdp.RData, originally downloaded from the npci repository - AMR-UTI dataset need to be obtained from PhysioNet, and place the
all_prescriptions.csv,all_uti_features.csv, andall_uti_resist_labels.csvfiles indata/AMR-UTIfolder.
pip install -r requirements.txt| Path | Description |
|---|---|
src/run_batch_main_sim_A.sh |
Run synthetic dataset A experiments method |
src/run_batch_main_sim_B.sh |
Run synthetic dataset B experiments usage |
src/run_batch_main_ihdp.sh |
Run IHDP experiments |
src/run_batch_main_amruti.sh |
Run AMR-UTI experiments |
| Path | Description |
|---|---|
src/analysis/variance_viz.ipynb |
Analyze asymptotic variance |
src/analysis/sim_experiment_results.ipynb |
Analyze synthetic datasets experiment results |
src/analysis/ihdp_experiment_results.ipynb |
Analyze IHDP experiments results |
src/analysis/amruti_experiment_results.ipynb |
AMR-UTI experiments results |
If you use LobsterNet in your research, please cite:
Chen W, Chang T, Wiens J. Conditional Front-door Adjustment for Heterogeneous Treatment Assignment Effect Estimation Under Non-adherence. Conference on Health, Inference and Learning (2025).
This project is licensed under the Apache 2.0 License.