A class of algorithms for general instrumental variable models

This is the Python code accompanying the paper

A Class of General Instrumental Variable Models
Niki Kilbertus, Matt J. Kusner, Ricardo Silva
Neural Information Processing Systems (NeurIPS) 2020

Setup

First clone this repository and navigate to the main directory

git clone [email protected]:nikikilbertus/general-iv-models.git
cd general-iv-models

To run the code, please first create a new Python3 environment (Python version >= 3.6 should work). For example, if you are using virtualenvwrapper run

mktmpenv -n

Then install the required packages into your newly created environment via

python -m pip install -r requirements.txt

Run experiments

There are three executable scripts in the general-iv-models directory to run different subsets of the experiments in the paper:

• linear_experiments.sh
• non_linear_experiments.sh
• sigmoid_design.sh

Running any of them will create a directory general-iv-models/results where all results (plots) will be stored.

To run all experiments, simply run

./linear_experiments.sh
./non_linear_experiments.sh
./sigmoid_design.sh