Distributional Instrumental Variable Method (DIV)

Distributional Instrumental Variable Method is a neural network-based method to estimate distributional causal effects (interventional distributions) proposed in the paper "Distributional Instrumental Variable Method" by A. Holovchak, S. Saengkyongam, N. Meinshausen and X. Shen (2025).

Installation

The latest release of the Python package can be installed through pip:

pip install div

Usage Example

from div import DIV

## Fit a DIV model to data (x, y, z)
model = DIV(1, 1, 1, num_layer=4)
model.train(x, y, z, 10000, print_every_iter=1000)

## Evaluation
x_eval = torch.linspace(x.min(), x.max(), 5000).unsqueeze(1)
## interventional mean function
y_est_mean = model.predict_causal(x_eval, sample_size=1000)

Contact information

If you meet any problems with the code, please submit an issue or contact Xinwei Shen.