Glossary

edge
relationship

An edge connects two nodes in a graph.

entity
node
vertex

The entities or objects in a graph, that are connected by edges.

iloc
integer location

Similar to ilocs in Pandas, ilocs in StellarGraph are sequential integers that allow for efficient storage and indexing. The stellargraph.StellarGraph class typically stores external IDs as ilocs internally.

inductive

A task is inductive if it generalises to unseen input. For node classification and link prediction tasks, this typically means a model trained on one graph (or subgraph) can be used for prediction on new nodes (whether a whole new graph or the larger graph that contains the training subgraph).

A task that predicts properties of edges in a graph. For example, predicting the probability of a “likes” edge between a user and a product (recommendation), or the strength of an atomic bond. This is typically framed as predicting new edges in a graph (where the predicted property is some measure of probability/likelihood of the edge existing), but includes predicting properties on existing edges.

See also

Demos of algorithms for link prediction.

node attribute inference
node classification

A task that predicts properties of individual nodes in a graph. For example, predicting the subject of an academic paper, or the flowering time of a crop. Despite the “classification” in the name, this includes regression tasks.

See also

Demos of algorithms for node classification.