This is the second education seminar on physics-guided machine learning. The focus is on model’s approximation, which is the core of ML implementation in physical sciences. Essentially, in ML, one simply approximates a model by projection onto an assumed vector space. The proper selection of the model space becomes accordingly the crucial step for proper “physics-guided” ML implementation. Today, many related approximation aspects will be discussed briefly such as model complexity, parsimony, the Rashoman’s set, and extrapolation vs. interpolation. As done in the last seminar, few real-life ML examples will be used for illustration.