neo4j-gds-skill

To run Neo4j GDS algorithms, you must first create a graph projection, then execute the algorithm using stream, stats, mutate, or write modes to control whether results are returned, stored in memory, or written back to the database.

FastRP embeddings and KNN similarity search in Neo4j require projecting the graph, generating the embeddings, and then running the KNN algorithm, often using mutate or write modes to save embeddings for later vector indexing.

Neo4j Graph Data Science requires the embedded GDS plugin with a graphdatascience Python client or direct gds.* Cypher procedures. It does not support Aura Graph Analytics sessions or AuraDB session-based projection workflows.

You can estimate memory in Neo4j GDS by running a pre-flight estimation procedure on your graph projection before executing large analytics workflows, helping you avoid out-of-memory errors and ensure safe execution.

Neo4j GDS algorithms fail with missing properties or out-of-memory errors when graph projections lack required node properties or when memory estimation limits are exceeded, requiring pre-flight checks and catalog cleanup.

Use stream mode for Neo4j graph data science algorithms to return results directly without side effects, and use write mode to persist algorithm results like PageRank scores back to the database for later querying.