What it does TypeDB-ML was a Python library that exported TypeDB knowledge graphs into formats usable by graph ML toolkits. It provided two main integrations: NetworkX for classical graph algorithms, and PyTorch Geometric for building Graph Neural Networks. The PyG path included a DataSet loader, feature encoders for typed data, and a worked link-prediction example.

The interesting bit The library tried to solve a genuinely fiddly problem: TypeDB’s strongly-typed, heterogeneous data maps naturally to PyG’s HeteroData, but PyG’s own conversion loses node ordering. TypeDB-ML added store_concepts_by_type to keep concepts and predictions properly aligned — a small but thoughtful piece of plumbing.

Key highlights

• Lazy graph loading from TypeDB into PyG Data objects
• FeatureEncoder with built-in encoders for continuous and categorical typed data
• NetworkX integration for algorithm libraries over exported graph queries
• Full link-prediction example in examples/diagnosis
• Bazel-based build and test system

Caveats

• Explicitly deprecated: README states “outdated and not supported” with planned closure by end of 2023
• Requires pinned dependency chain: Python 3.7+, TypeDB 2.11.1, typedb-client-python 2.11.x — versions now frozen in time
• PyG installation noted as finicky even when actively maintained

Verdict Worth studying if you’re building similar graph-to-ML bridges and want to see how someone else handled type-aware heterogeneous conversion. Not worth installing unless you’re already maintaining a legacy TypeDB 2.11 deployment and need the specific example code.