causaLens logocausaLens logo alternate

Unsuitability of NOTEARS for Causal Graph Discovery

The field of Causal Discovery aims to identify causal relationships from purely observational data. In this paper, causaLens demonstrates that NOTEARS — a popular causal discovery method — is not suitable for identifying truly causal relationships from data.

SCROLL DOWN

Download our research paper

This paper proves that NOTEARS lacks scale-invariance by showing how a simple rescaling of the variables can lead to a large alteration of the derived DAG.