Python APIs for causal modeling algorithms developed by the University of Pittsburgh/Carnegie Mellon University Center for Causal Discovery.

This code is distributed under the LGPL 2.1 license.


Python 2.7 and 3.6

  • javabridge>=1.0.11
  • pandas
  • numpy
  • JDK 1.8
  • pydot (Optional)
  • GraphViz (Optional)

Docker Image

A pre-installed py-causal Docker image is available at Docker Hub

Installation overview:

To install on existing Python installation, we have found two approaches to be useful: Direct python installation with pip, possibly including use of Jupyter. This approach is likely best for users who have Python installed and are familiar with installing Python modules. Installation via Anaconda, which installs Python and related utilities.

Directions for both approaches are given below...

Installation with pip

If you do not have pip installed already, try these instructions.

Once pip is installed, execute these commands

pip install -U numpy
pip install -U pandas
pip install -U javabridge
pip install -U pydot 
pip install -U GraphViz

Note: you also need to install the GraphViz engine by following these instructions.

We have observed that on some OS X installations, pydot may provide the following response Couldn't import dot_parser, loading of dot files will not be possible.

If you see this, try the following

 pip uninstall pydot
 pip install pyparsing==1.5.7
 pip install pydot

Then, from within the py-causal directory, run the following command:

python install

or use the pip command:

pip install git+git://

After running this command, enter a python shell and attempt the follwing import import pandas as pd import pydot from tetrad import search as s

Finally, try to run the python example


Be sure to run this from within the py-causal directory.

This program will create a file named tetrad.svg, which should be viewable in any SVG capable program. If you see a causal graph, everything is working correctly.

Running Jupyter/IPython

We have found Jupyter notebooks to be helpful. (Those who have run IPython in the past should know that Jupyter is simply a new name for IPython). To add Jupyter to your completed python install, simply run

pip -U jupyter
jupyter notebook

and then load one of the Jupyter notebooks found in this installation.


Installing Python with Anaconda and Jupyter may be easier for some users:

For OS X, this default install does not seem to work well. try the following instead:

conda install --channel python-javabridge

Then run the following to configure anacoda

conda install pandas  
conda install numpy
conda install pydot
conda install graphviz 
conda install -c pycausal 
jupyter notebook

and then load one of the Jupyter notebooks.