Before building Graph Neural Networks, we must first establish an understanding of the data they operate on: graphs. This chapter addresses the fundamentals of working with graph-structured data in a machine learning context.

We will begin by defining what constitutes a graph and examining why conventional neural networks, like CNNs and RNNs, are ill-suited for this data type. Following this, we will cover the primary machine learning tasks performed on graphs, such as node classification, link prediction, and graph classification.

A significant portion of this chapter is dedicated to the practical matter of representing graphs for computation. You will learn how to encode a graph's structure and attributes using standard formats like adjacency matrices ($A$) and node feature matrices ($X$). To conclude, we will put these ideas into practice with a brief introduction to the NetworkX library, using it to load and inspect a graph dataset.