If you strive to get the causal model right...

...you've got a smaller chance of getting your data science problems wrong.

Hello, fellow datanistas!

As promised last month, I wanted to share some resources on causal modelling for this month's newsletter. The past decade of data science has shown that predictive models may be powerful. But it's pretty well known that if we get the causal structure of a model correct, we more often than not get predictive power for free. When I dug into causal inference for the first time, it was an eye-opener, and I wanted to share some resources that really helped me along the way.

Lectures by Jonas Peters

Jonas Peters' lectures on causality were my first introduction to the topic of inferring causal structure from observational data, a.k.a. causal inference. For my own convenience, I collated them together into a YouTube playlist. In studying these lectures, I basically took notes while watching them. As I'll share at the end, my notes culminated in a collection of Jupyter notebooks that helped me really make ideas concrete.

Causal Salad, Design, and Inference

Richard McElreath writes about these three ways of doing science on his blog, Elements of Evolutionary Anthropology. Part 2, which is all about Causal Design, provides a beautiful introduction to causal thinking, which really begins with coming up with a generative model of our observed data.

Causal Data Science Essays

Adam Kelleher has a great series that I point many to for learning about causal inference and how to use them in a data science context. I believe it was from Adam where I found the words best expressed - if we solve the causal problem, we solve the prediction problem for free.

DoWhy: A Package for Causal Inference in Python

Amit Sharma and colleagues at Microsoft released a package, DoWhy, that helps Python users apply causal inference methods to their own problems. The docs and repository are all open-source!

From my collection

As I mentioned above, I created a collection of Jupyter notebooks to make the ideas I learned more concrete. Learning in the open has been my thing, and I'm not ashamed to share where I am in this learning journey with everyone!

For those interested in Causal Inference, I hope this special newsletter edition brought you something valuable. Please forward it to your colleagues and share the love and knowledge around if you found it helpful!

Stay safe, stay hacking!