Installation

You can install the package from PyPI with the command pip install medil. It’s recommended to install in a fresh virtual environment, for example using pipenv, in which case the command becomes pipenv install medil. ..

The default only requires SciPy and allows for learning causal factor models from data, random generation of causal factor models, and sampling data from a given (or learned) causal factor model, all in the linear Gaussian setting. There are additional optional requirements for extra features:

Key	Description
all	Installs all optional dependencies, for full functionality; equivalent to using all of the other keys below.
dgm	Uses deep generative models for the causal mechanisms (instead of restricting to the linear Gaussian setting), specifically using PyTorch to implement a variational autoencoder.
vis	Uses NetworkX, Matplotlib and seaborn for vizualize the causal factor graph and various plots.

In order to use these features, simply append the corresponding comma-separated keys after the package name in brackets (with no spaces), e.g., pip install medil[vis].