What problem does it solve? Running graph algorithms like PageRank, Louvain, or FastRP against Neo4j requires correct graph projection, memory estimation, execution mode selection, and version-aware API usage — mistakes cause OOM errors, failed procedures, or silently wrong results. This Skill provides verified patterns for the embedded GDS plugin so agents produce working code on the first attempt. ## Core Features & Use Cases - Graph Projection & Catalog Management: Native and Cypher projection, undirected orientation, subgraph filtering, memory estimation, and graph drop/list operations. - Algorithm Execution: PageRank, Louvain, Leiden, WCC, Betweenness Centrality, Node Similarity, FastRP embeddings, and KNN across stream/stats/mutate/write modes, with v2 Python client endpoints preferred and v1 fallbacks labeled. - Recommendation Pipelines: The FastRP → KNN pattern builds similarity relationships from node embeddings, ready for Neo4j vector indexes. - Use Case: A data engineer needs product recommendations from a co-purchase graph. The Skill projects the graph undirected, estimates memory, mutates FastRP embeddings, writes KNN similarity relationships, and drops the projection. ## Quick Start Ask the agent to project a named graph from your Neo4j database and run PageRank on it using the GDS Python client.
