This repository is extended from https://github.com/authors-1901-10912/A-Meta-Transfer-Objective-For-Learning-To-Disentangle-Causal-Mechanisms.
To avoid any conflict with your existing Python setup, and to keep this project self-contained, it is suggested to work in a virtual environment with virtualenv. To install virtualenv:
pip install --upgrade virtualenv
Create a virtual environment, activate it and install the requirements in requirements.txt.
virtualenv venv
source venv/bin/activate
pip install -r requirements.txt
Causality direction prediction
cd comparison_experiments/direction_prediction
python baseline.py
python proposed.py
python plots.py
Representation learning
cd comparison_experiments/representation_learning
python3 baseline.py
python3 proposed.py
python3 plots.py
Robustness
cd comparison_experiments/counter_example_discrete
python baseline.py
python proposed.py
Causality direction prediction with N=100
cd comparison_experiments/direction_prediction/
python baseline.py --N 100
python proposed.py --N 100
python plots_N=100.py
Causality direction prediction with continuous variable
cd comparison_experiments/direction_prediction_continuous
python baseline.py
python proposed.py
Other metrics
cd comparison_experiments/other_metrics
python kl_divergence.py
python grad_l2_norm.py