Avik Basu - Beyond Just Prediction: Causal Thinking in Machine Learning | PyData Seattle 2025 #machinelearning #correlation #causality #uplift #metalearner #slearner #tlearner #xlearner #causalml #scikituplift #econml

My PR to the #EconML #PyWhy #opensource #causalai project was merged! 🎉 I made a small contribution by allowing a flexible choice of evaluation metric for scoring both the first stage and final stage models in Double Machine Learning (#DML).
#CausalInference #machinelearning #datascience

oh one other change: econml now (rightfully) distinguishes between covariates used to fit E[Y|X], E[T|X] and the ones used to estimate CATE functions. so i got it to run by passing the same matrix twice (i.e. hetfx by all covars - grf default) but that is probably not what anyone should ever do

The LOST-STATS page on causal forest has a Python implementation that apparently no longer works. Anyone familiar with econml and able to help update the code? I can handle the page update if you can fix the code lost-stats.github.io/Machine_Lear...

Made my first PR on an #OpenSource project. Before now, all the projects I worked with had everything I needed. The PR is just a small suggestion to improve memory performance in #EconML (#causalml) - we'll see what the repo maintainers think of my hackery, haha...

My journey was something like this

1. Jonas Peters introductory lectures
2. Judeal Pearl's Book of Why, causality and Primer
3. Rubin's PO, Angrist, Imbens work on IV
4. Hernan and Robin's What If book
5. Coding causal models with R/Python (dagitty, doWhy, econML)
6. Causality in LLMs

#CausalSky

As a default, it seems reasonable to go w/ DoubleML if you're sure that DML is a methodology you want to use & there's a specific estimator you want to use impl. in the pkg

EconML seems a better choice when you're experimenting with various estimators &/or want access to the DoWhy/EconML feats

3/3

...evaluation module and more.

DoubleML has more use case specific implementations of the DML framework, a great sensitivity analysis module (a feat. also available in EconML, but less advanced (at least this is how it was some time ago)), and more.

2/n

Hi Jake,

Yes, I am. EconML is a part of a broader ecosystem and it allows you to easily compare different types of estimators (not only DML). DoubleML focuses on DML specifically.

EconML gives you access to all the functionalities of the DoWhy/EconML ecosystem, including refutation tests,...

1/n

you don't know how strongly i agree with you; spending weeks F12-ing econml classes to figure out where the linalg is done is excellent anti-OOP indoctrination. DRY principle is good, but within reason.

Introductory post! We are causalwizard.app , a website that helps researchers learn about and adopt Causal Inference. The app has a database of about 100 practical causality articles and includes tools from DoWhy, EconML and StatsModels.
#CausalSky #EpiSky #MachineLearning #DataScience #Statistics

Anyone know of anything like "A list of Python packages an applied economist should install to allow Python to do what stata does" - just a nice list of all the great packages people have made to get estimation tools in Python. I'm thinking e.g. pyfixest, binsreg, econml, etc. Any leads?

The new version of EconML offers some cool new features (stay tuned, we'll cover them soon) and a couple of very interesting code bases came to life with recent NeurIPS papers (we'll discuss some of them this year as well).

3/n