Translator._parse_memgraph uses get_edges_from_db, which matches all edges in Memgraph. For large graphs this is slow and possibly unnecessary (like for my use-case).
I've rewritten the Translator and DGLTranslator classes for my use-case, by adding a Match parameter in .get_instance which is propagated down to the get_edges_from_db function. By providing a Match query on node labels and relationship properties, I can use the customised DGLTranslator to return subgraphs as part of a PyTorch/DGL Dataloader - making it very convenient for model training/prediction.
Is there a technical/conceptual reason the existing Translator class doesn't support optionally fetching a subgraph? My solution is specific to the use-case I have, and as-is, it definitely isn't ready for a PR. But if there is interest for this kind of functionality, I'd be happy to contribute.
Translator._parse_memgraphusesget_edges_from_db, which matches all edges in Memgraph. For large graphs this is slow and possibly unnecessary (like for my use-case).I've rewritten the
TranslatorandDGLTranslatorclasses for my use-case, by adding a Match parameter in.get_instancewhich is propagated down to theget_edges_from_dbfunction. By providing a Match query on node labels and relationship properties, I can use the customised DGLTranslator to return subgraphs as part of a PyTorch/DGL Dataloader - making it very convenient for model training/prediction.Is there a technical/conceptual reason the existing Translator class doesn't support optionally fetching a subgraph? My solution is specific to the use-case I have, and as-is, it definitely isn't ready for a PR. But if there is interest for this kind of functionality, I'd be happy to contribute.