search

munichpavel / risk-ai-workshop

Lecture material and sample code for the workshop "Risk, Artificial Intelligence and Discrete Geometry" at the University of Ljubljana
GNU General Public License v3.0
7 stars 8 forks source link

Fix average-treatment-effect do calculation in correlation-causality slides #20

Closed munichpavel closed 2 years ago

munichpavel commented 2 years ago

In slides of 97f0b49b1fdcd687c1f814d1d70802d0d05972ca, e.g., the historical and do versions of the average treatment effect probabilities look the same, which contradicts the claim that the do version has half of the ATE of the historical ones on the slide from 2 years ago, e.g. here:

https://github.com/munichpavel/risk-ai-workshop/tree/fe4dbfa65fef19d47a48f026178fd2bb678c0865/slides/correlation-causality

Fix this.

munichpavel commented 2 years ago

Root cause

was a refactoring from

https://github.com/munichpavel/risk-learning-exercises/commit/41403f014889b2783e96abea5b2744d8d519034b

munichpavel commented 2 years ago

Fixed at e06e4bdd295ffec3f5c2509c9ffecbad61b57f4f; see also (if you have access) https://github.com/munichpavel/risk-learning-exercises/tree/3db5383eb751690b157e56d6daa712608368c83d