This project implements the TARNet, CFRNet and DragonNet models using Tensorflow with an exploration on ablation studies and fine-tuning on hyperparameters.

This is Tensorflow 2.8.0 implementation of the following models: TARNet, CFRNet and DragonNet based on the following papers:

1. Estimating individual treatment effect: generalization bounds and algorithms. Uri Shalit, Uri Shalit, Fredrik D. Johansson, David Sontag PMLR 2017 [PDF]

2. Adapting Neural Networks for the Estimation of Treatment Effects. Claudia Shi, David M. Blei, Victor Veitch NIPS 2019 [PDF]

3. Deep Learning of Potential Outcomes. Bernard Koch, Tim Sainburg2, Pablo Geraldo Bastias, Song Jiang, Yizhou Sun, Jacob Foster SocArXiv [PDF]

Check the packages needed or simply run the command Requirements

• tensorflow==2.8.0
• scikit-learn==0.24.2
• numpy==1.21.5
• pandas==1.3.4
• keras-tuner==1.0.4

❱❱❱ pip install -r requirements.txt

.
├── README.md
├── data
│   ├── IHDP
│   │   ├── ihdp_npci_1-100.test.npz
│   │   └── ihdp_npci_1-100.train.npz
│   └── SIPP
│       └── sipp1991.dta
├── main.py
├── model
│   ├── common_layer.py
│   ├── model_loss.py
│   ├── model_metrics.py
│   └── models.py
├── notebook
│   ├── causalDL_hyperparam_opitm.ipynb
│   └── notebook_test.ipynb
├── requirements.txt
├── save

Data is located in folder ./data/IHDP.

IHDP dataset is a semi-synthetic dataset based on a randomized experiment of Infant Health and Development Program. A more detailed introduction to this dataset can be found at
https://www.researchgate.net/publication/11523952_Infant_Mortality_Statistics_from_the_1999_Period_Linked_BirthInfant_Death_Data_Set
and available for download at
http://www.fredjo.com/files/ihdp_npci_1-100.train.npz
http://www.fredjo.com/files/ihdp_npci_1-100.test.npz

Example of command: TARNet

❱❱❱ python3 main.py --model tarnet --dataset IHDP

CFRNet

❱❱❱ python3 main.py --model cfrnet --dataset IHDP

DragonNet (with AIPW)

❱❱❱ python3 main.py --model dragonnet --dataset IHDP

DragonNet (with Targeted Regularization)

❱❱❱ python3 main.py --model dragonnetTR --dataset IHDP

• Ablation

• ITE

• ITE with fine-tuning

Shaoyu Liu & Jinhui(Ferry) Wu