PIPCFR: Pseudo-outcome Imputation with Post-treatment Variables for Individual Treatment Effect Estimation

Installation

You will need to install:

• geomloss==0.2.4
• numpy
• pandas
• scikit_learn==1.0.2
• scikit_uplift==0.4.0
• torch==1.10.1
• tensorboard
• protobuf==3.20
• scikit-optimize
• ot==0.9.5
• tensorflow-gpu==1.15.0
• causalml

The dependencies for Pairnet and ESCFR from github:

git clone https://github.com/nlokeshiisc/pairnet_release.git

Generate Training Data

IHDP

cd data/IHDP
python generate_data.py
cd ..

NEWS

cd data/NEWS
python generate_data.py
cd ..

Synthetic Data

cd data/Synthetic
bash generate.sh
cd ..

Run Experiments

Run experiments for NEWS dataset

MODELS_JSON='./examples/model_news.json' python run_experiment.py

Run experiments for IHDP dataset

MODELS_JSON='./examples/model_ihdp.json' python run_experiment.py

Run experiments for Synthetic dataset

MODELS_JSON='./examples/model_synthetic.json' python run_experiment.py

MODELS_JSON is the configure information of models. You can add model in ./examples/.

Citation

If you find this project useful in your research, please consider citing:

License

• PIPCFR is licensed under the Apache License 2.0.