Causal discovery – assumptions refresher
The first time we mentioned causal discovery in this book was in Chapter 1. In Chapter 5, we went a little bit deeper and discussed two assumptions that are often used for causal discovery methods: faithfulness and minimality.
In this section, we’ll review these assumptions and discuss other, more general assumptions that will be useful in our causal discovery journey.
Let’s start!
Gearing up
Causal discovery aims at discovering (or learning) the true causal graph from observational (and sometimes interventional or mixed) data.
In general, this task is difficult but possible under certain conditions. Many causal discovery methods will require that we meet a set of assumptions in order to use them properly.
The first general assumption is one of causal sufficiency (or lack of hidden confounding). A vast majority of causal discovery methods rely on this assumption (although not all).
Another popular assumption...
