This repository contains the code for the paper "Estimating causal effects using neural autoregressive density estimators". The repository is composed of 11 scripts:

• ./data/fake_data.py
• ./experiments/default_experiment_yaml.py
• ./bootstrap.py
• ./hyperparameter_search.py
• ./main.py
• ./src/models/causal_estimates.py
• ./src/models/data_loader.py
• ./src/models/model.py
• ./src/models/train.py
• ./src/utils/plot_utils.py
• ./src/utils/utils.py

In order to run the experiments, we need to generate the data and run a command line tool. The process is the following:

To generate fake data, run in the command line: python3 ./data/fake_data.py n path Where n is the number of samples you want to generate and path is the folder where you want to save the fake data.

In order to run an experiment, you need a YAML file with the parameters. In the "experiments" folder you can find a tool to create a default parameters YAML. In order to run that script, run in the command line python3 ./experiments/default_experiment_yaml.py dir where dir is the directory where you want to save the parameters YAML.

Having the data, and the YAML parameters file, we can run an experiment by running python3 main.py data_dir yaml_dir. The results will be recorded in the "results" directory. You can also perform a hyper-parameter search with: python3 hyper_parameter.py data_dir yaml_dir, or a bootstrap estimate of a particular model with: python3 bootstrap.py data_dir yaml_dir.

In order to cite this code or the paper, use the following bib:

@article{garrido2020estimating,
  title={Estimating Causal Effects with the Neural Autoregressive Density Estimator},
  author={Garrido, Sergio and Borysov, Stanislav S and Rich, Jeppe and Pereira, Francisco C},
  journal={arXiv preprint arXiv:2008.07283},
  year={2020}
}