Information Measures on Causal Graphs.

## Project description

causalinfo is a Python library to aid in experimenting with different information measures on causal graphs—a combination of information theory with recent work on causal graphs [Pearl2000]. These information measures can used to ascertain the degree to which one variable controls or explains other variables in the graph. The use of these measures has important connections to work on causal explanation in philosophy of science, and to understanding information processing in biological networks.

The library is a work in progress, and will be extended as research continues.

## What does it do?

causalinfo has been written primarily for interactive use within IPython Notebook. You can create variables and assign probability distributions to them, or relate them to other variables using conditional probabilities. Several related variables can be combined into a directed acyclic graph, which can generate a joint distribution for all variables under observation, or under controlled interventions on certain variables. You can also calculate various information measures between variables in the graph whilst controlling other variables. These include correlative measures, such as Mutual Information, but also causal measures, such as Information Flow [AyPolani2008], and Causal Specificity [GriffithsEtAl2015].

For some brief examples of how to use the library, please see the IPython Notebooks that are included:

• Introduction. A short introduction to some of the things you can do with the library.

• Rain. Performing some interventions on a causal graph; an example from Judea Pearl’s book.

• Signaling. Measuring Causation in Signaling Networks. Some examples from [CalcottEtAl2016].

• Information Flow. Measuring the flow of information in Causal networks from [AyPolani2008].

## Some Caveats

The library is not meant for large scale analysis. The code has been written to offload as much as possible on to other libraries (such as Pandas and Networkx), and to allow easy inspection of what is going on within IPython Notebook, thus it is not optimized for speed. Calculating the joint distribution for a causal graph with many variables can become very slow (especially if the variables have many states).

## Authorship

All code was written by Brett Calcott.

## Acknowledgments

This work is part of the research project on the Causal Foundations of Biological Information at the University of Sydney, Australia. The work was made possible through the support of a grant from the Templeton World Charity Foundation. The opinions expressed are those of the author and do not necessarily reflect the views of the Templeton World Charity Foundation.

## References

[AyPolani2008] (1,2)

Ay, N., & Polani, D. (2008). Information flows in causal networks. Advances in Complex Systems, 11(01), 17–41.

Griffiths, P. E., Pocheville, A., Calcott, B., Stotz, K., Kim, H., & Knight, R. (2015). Measuring Causal Specificity. Philosophy of Science, 82(October), 529–555.

Calcott, B., Griffiths, P. E., Pocheville, A. (Forthcoming). Signals that Make a Difference. British Journal for Philosophy of Science.

Pearl, J. (2000). Causality. Cambridge University Press.

## Project details

Uploaded source