Graph dataset split

how is the data set of graph type divided into test set and training set? Does dividing a graph into training set and test set not change the structure of the graph or affect the model effect to fit the data?(such as cora dataset)

1 Like

It depends on the learning task as well as whether the scenario is transductive or inductive. For example, transductive node classification will use a mask tensor to indicate whether a node belongs to training, validation or test set. During training, the model will have the access to the entire graph structure but only the labels of the training nodes are used. By contrast, for inductive node classification, the model will be typically trained on one graph but tested on another. In this case, the model need to generalize across different graph structures. Edge classification or link prediction is more complicated. You may want to take a look at the tutorials here:

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.