If machine learning is cool, how about double machine learning?
The title of this section is inspired by Brady Neal’s video (https://bit.ly/BradyNealDML) where he (jokingly) suggests that DML “is maybe twice as cool” as regular machine learning. Let’s see!
DML, also known as debiased machine learning or orthogonal machine learning, is another causal framework with the root double in its name. In DML – similarly to DR methods – we fit two models that estimate different parts of the relationships in the data.
DML can be implemented using arbitrary base estimators, and in this sense, it also belongs to the meta-learner family. At the same time, unlike S-, T- and X-Learners, the framework comes with a strong theoretical background and unique architectural solutions.
In this section, we’ll introduce the main concepts behind DML. After that, we’ll apply it to our earnings dataset using DoWhy’s API. We’ll discuss...
