Neo4j GDS Certification flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

PageRank damping factor

The probability a random surfer continues along an edge rather than teleporting; default 0.85. Higher values weight link structure more heavily and lower values flatten scores toward uniform.

Betweenness centrality

For each node, the fraction of all shortest paths between other node pairs that pass through it. High scores mark brokers or bridges; it is expensive, so sampling is often used.

Closeness centrality

Average of the shortest-path distances from a node to all other reachable nodes, inverted so higher means closer. Nodes in disconnected components get scores based only on reachable nodes.

Louvain modularity

A community detection algorithm that greedily maximizes modularity by local node moves then aggregates communities into super-nodes and repeats until no gain is possible.

Leiden algorithm

An improvement on Louvain that adds a refinement phase, guaranteeing internally connected communities and often producing higher-quality, more stable partitions on the same graph.

Dijkstra shortest path

Finds minimum-cost paths from a source using relationshipWeightProperty; all weights must be non-negative. Returns path, nodeIds, and totalCost per target.

A* algorithm

Dijkstra with a geographic heuristic using latitude and longitude node properties, reducing explored nodes for point-to-point routing. Requires the same non-negative weight constraint.

Node similarity Jaccard

Compares two nodes by the overlap of their neighbor sets: intersection size divided by union size. similarityCutoff filters weak pairs and topK limits results per node.

kNN cosine similarity

Computes cosine similarity between node property vectors, returning topK nearest neighbors per node. Suited to embeddings where neighbor overlap is not the right signal.

Label propagation

Iteratively assigns each node the most frequent label among its neighbors until stable. Very fast and scalable, but results can vary with node ordering and concurrency.

Triangle count

Counts triangles each node participates in, indicating local clustering. The local clustering coefficient normalizes by possible pairs among the node's neighbors.

Stream mode

Algorithm execution mode that returns result rows to the client without modifying the graph or the projection. Used for inspection and small result sets.

Mutate mode

Execution mode that writes algorithm results as node or relationship properties into the in-memory graph projection only. Fast for chaining algorithms without database writes.

Write mode

Execution mode that persists algorithm results as properties on nodes or relationships in the Neo4j database, in addition to returning summary statistics.

Graph projection creation

gds.graph.project defines a named in-memory graph from node and relationship projections, optionally with orientation and property selections, for later algorithm calls.

Native projection

Projection built directly from stored node labels and relationship types, selecting properties up front. Faster than Cypher projection and the default choice for large graphs.

Cypher projection

Projection defined by nodeQuery and relationshipQuery strings that must return id plus source and target columns. Flexible for filtered or computed subgraphs at higher cost.

UNDIRECTED orientation

Relationship projection option that treats each stored relationship as traversable in both directions, doubling effective degree and changing centrality and community results.

Memory estimation

Calling write.estimate or mutate.estimate before execution returns requiredMemory and bytesMin/bytesMax, letting you size the GDS heap before an algorithm runs.

gds.graph.list

Returns all current graph projections with node and relationship counts, memory usage, and configuration. Used to audit projections before dropping or reusing them.

gds.graph.drop

Removes a named graph projection and frees its memory. Projections do not survive restart, so dropping unused graphs is the primary remedy for heap pressure.

Job cancellation

gds.beta.listProgress lists running jobs with jobId; gds.beta.job.cancel stops a job by id. Useful when a long algorithm run is misconfigured.

Link prediction pipeline

A trained model workflow created with gds.beta.pipeline.linkPrediction.create, combining node features and a classifier to score probable missing relationships.

Plugin installation

Place the gds JAR in the Neo4j plugins directory and allowlist gds.* in dbms.security.procedures.allowlist and unrestricted settings, then restart the database.

gds.version

Procedure returning the installed GDS library version string. Confirms the plugin loaded correctly and which algorithm namespaces are available.

gds.util.nodeProperty

Helper that reads a projected node property by node id without streaming the entire graph. Handy for post-processing algorithm output in Cypher.

GDS Enterprise license

Commercial license configured via gds.enterprise.license_file or the gds.license setting. Unlocks the full algorithm set beyond the Community edition limits.

gds.beta namespace

Prefix for beta procedures such as pipelines and model management. Functionality is stable enough for use but may change across minor releases.

gds.alpha namespace

Prefix for experimental procedures. Behavior, configuration, and names can change or be removed between releases, so avoid depending on them in production.

Anonymous graph

A graph created implicitly by passing a node or relationship specification directly to an algorithm call, not stored in the catalog and not reusable.

GDS heap sizing

GDS uses off-heap memory for projections in addition to Neo4j heap. Under-provisioning causes OutOfMemory during projection or algorithm execution.

Graph catalog scope

Projections are per-database catalog objects held in memory, invisible to other databases, and lost on restart, requiring re-projection after every restart.