Neo4j GDS Certification study plan
Neo4j GDS Certification · Give each week a clear focus.
8 weeks · About 7 hours per week
Algorithms
Map the GDS algorithm catalog into the four families (pathfinding, centrality, community, similarity) and label each algorithm's execution mode (stream, stats, mutate, write)
Configure a local Neo4j instance with the GDS plugin and project a named in-memory graph from a sample dataset using gds.graph.project
Run PageRank and betweenness centrality in stats mode and interpret the returned centrality distribution percentiles
Explain the difference between nodeProperty and relationshipWeightProperty parameters and when dampening factors matter
Your checkpoint
A one-page cheat sheet listing every algorithm family with its required parameters and execution modes, plus a screenshot-free written summary of PageRank vs betweenness results on the same graph
Algorithms
Implement a weighted shortest-path query using Dijkstra with relationshipWeightProperty and compare results against the unweighted BFS variant
Configure A* with latitude/longitude node properties as the heuristic to speed up route queries
Demonstrate single-source shortest path and all-pairs shortest path on a small network and explain the computational cost difference
Tune maxIterations, tolerance, and relationshipWeightProperty defaults for convergence-sensitive algorithms
Your checkpoint
A written comparison table of Dijkstra, A*, and BFS outputs on the same 20-node network, including runtime observations and one paragraph on when each should be chosen
Algorithms
Run Louvain and Label Propagation in mutate mode and inspect the modularity and community count in the mutate summary
Write back community IDs with gds.louvain.write and validate community cohesion using a Cypher aggregation query
Compare Leiden against Louvain on the same graph and articulate the refinement step that differs
Explain the role of relationshipWeightProperty and the maxLevels parameter in hierarchical community detection
Your checkpoint
A community detection report on a sample social graph showing modularity scores for Louvain and Leiden, community size distribution, and a justified recommendation of which algorithm to use
Algorithms
Execute node similarity (Jaccard and Overlap) in stream mode and filter results with a similarityCutoff
Run KNN and K-Nearest Neighbors with the same cutoff and explain the difference between the two similarity approaches
Interpret the results of a link prediction pipeline using the common neighbors and Adamic-Adar predictors
Diagnose the memory estimate returned by gds.nodeSimilarity.stats and adjust concurrency accordingly
Your checkpoint
A similarity analysis write-up showing top-10 similar node pairs under Jaccard, Overlap, and KNN, with a comparison of result counts and a note on memory estimation
Algorithms & Workflows
Build an end-to-end workflow: project a graph, run a centrality algorithm in mutate mode, then use the mutated property in a downstream community detection call
Use gds.graph.nodeProperty.stream and gds.graph.relationshipProperty.stream to inspect in-memory graph contents
Chain mutate results across two algorithms without writing to the database and explain why this is more efficient
Clean up projected graphs with gds.graph.drop and verify no orphaned graphs remain via gds.graph.list
Your checkpoint
A runnable Cypher script that projects, mutates, chains two algorithms, and drops the graph, saved with expected output annotations
Workflows
Construct a graph projection using Cypher projection syntax with node and relationship queries, including property selection
Use native projection with multiple node labels and relationship types and verify counts against the source database
Apply gds.graph.export to write a projected graph to a new database and confirm the exported schema
Estimate projection memory with gds.graph.project.estimate before running the real projection
Your checkpoint
A projection cookbook containing three working projection examples (native single-label, native multi-label, Cypher-based) with memory estimates and node/relationship counts verified against source queries
Workflows & GDS Library
Tune algorithm execution with concurrency, writeConcurrency, and the jobId/ job management procedures
Implement a workflow using gds.beta.pipeline.nodeClassification with train/test splits and feature steps
Configure node property steps and feature pipelines, then train a model and inspect model metrics
Explain GDS library components: the plugin architecture, the system database procedures, and the licensing/edition constraints on production features
Your checkpoint
A machine learning pipeline notebook-style script that projects a graph, builds a feature pipeline, trains a node classification model, and prints accuracy metrics, plus a short note on edition constraints
Algorithms & Workflows & GDS Library
Complete a timed 60-question mixed-domain practice set and score above 80 percent
Review every missed question by re-reading the relevant algorithm or workflow documentation and writing a corrected explanation
Re-run two previously built workflows from memory without consulting notes to confirm retention
Consolidate a final one-page reference of algorithm families, execution modes, and projection syntax for last-minute review
Your checkpoint
A scored practice exam log with per-domain accuracy, a list of corrected misconceptions, and the final consolidated reference sheet
Milestones
Week 2: Can explain and correctly parameterize Dijkstra, A*, BFS, and SSSP without notes, and justify weighted versus unweighted choices
Week 4: Can run at least four centrality and three community/similarity algorithms in the correct execution mode and interpret their output metrics
Week 6: Can build native and Cypher projections from scratch, estimate memory, and chain mutate results across two algorithms in a single workflow
Week 8: Scores at least 80 percent on a timed mixed-domain practice set and can articulate the corrected reasoning for every missed item