Graph Models

Use graph models when relationship structure is the thing you want to test. That usually means directed origin-to-destination flow, repeated route markets, neighborhood structure, or link likelihood.

Use CartoBoost Boosting Model Guides for row-level trees and CartoBoost Neural Model Guides when repeated IDs should be embedded without an explicit graph.

Choose A Guide

Guide	Best when
CartoBoost Node2Vec Graph Models	Topology and flow patterns matter more than node attributes.
CartoBoost GraphSAGE Models	Node attributes should shape the learned representation.
CartoBoost HeteroGraphSAGE Models	Relation IDs matter, but a single node-feature table is enough.
CartoBoost HinSAGE Models	Node types and valid source-relation-target combinations must be enforced.

Keep direction explicit. A source-to-target fact is usually not the same as the reverse fact.

See Graph Features for input shapes, directionality, generated features, saving and loading, and common failure modes.