Graph Classification (Supervised)

Graph classification is similar to the node classification formulation. In a graph classification task, we predict the attribute of a graph, unlike the label of a node (as done in the node classification task). Each graph is labeled, and, depending on the number of labels, it can be formulated as a binary or multiclass classification task. Depending on the use case, we can even formulate it as a regression task in which the target variable is a continuous value.

In the illustration above, we have four different graphs, out of which two graphs have similar attributes (shown in yellow or red). The idea is to classify the G5 graph using the knowledge of other graph attributes. In this instance, this can be formulated as a binary classification task.

Let's look at an example of the Reddit 10K ...