neo4j-graph-data-science-2.10.1

neo4j-graph-data-science-2.10.1

Breaking changes

• Raise an exception if relationship types are specified in undirectedRelationshipTypes or inverseIndexedRelationshipTypes, but not projected in the graph projection. This affects the following projections:
  • Cypher projection
  • Apache Arrow projection

New features

• Added procedures RandomWalk mutate mode:
  • gds.randomWalk.mutate
  • gds.randomWalk.mutate.estimate

Bug fixes

• Modified the FastRP initialization procedure to scale both the random values and property values to match the range of the embedding dimension. This adjustment addresses the issue of discontinuous behavior of embedding values when propertyRatio is close to 0 or 1.
• Fixed a bug where NodeSimilarity using relationship weights and JACCARD similarity metric produced incorrect similarity values.
• Fixed a bug where Graph implementations that do not override streamRelationships could not export relationships with a single property via arrow.
• Fixed a bug where Cypher projections would fail when a string, number or boolean literal is used in the query when query literal obfuscation is turned on.

Improvements

• Added failIfMissing configuration option to gds.graph.nodeProperties.drop procedure.
• Added progress logging for the gds.graph.project aggregation function.

Other changes

• The arrow triplet import action now accepts job_id and task_volume parameters.

2.9.0

New features

• Added a procedure for finding bridges in a graph
  • gds.bridges.stream
• Added procedures for finding articulation points in a graph
  • gds.articulationPoints.mutate
  • gds.articulationPoints.mutate.estimate
  • gds.articulationPoints.stats
  • gds.articulationPoints.stats.estimate
  • gds.articulationPoints.stream
  • gds.articulationPoints.stream.estimate
  • gds.articulationPoints.write
  • gds.articulationPoints.write.estimate

Bug fixes

• Fix a bug introduced in 2.7.0 where gds.wcc.mutate would ignore the consecutiveIds configuration parameter.
• Fixed a bug in 2.8.0 which caused errors writing back relationships to Neo4j. This affected these procedures:
  • gds.nodeSimilarity.filtered.write

Graph Data Science 2.8.0

Bug fixes

• Fixed a bug which caused writing back relationships to Neo4j potentially resulting in deadlocks.
  This affected these procedures:
  • gds.knn.write
  • gds.knn.filtered.write
  • gds.nodeSimilarity.write
  • gds.nodeSimilarity.filtered.write
  • gds.steinerTree.write
  • gds.spanningTree.write
  • gds.allShortestPaths.delta.write
  • gds.shortestPath.dijkstra.write
  • gds.shortestPath.astar.write
  • gds.shortestPath.yens.write
  • gds.bellmanFord.write
  • gds.graph.relationships.write
  • gds.graph.relationshipProperties.write

Other changes

• Reverted parallel relationship write-back introduced in 2.7, due to deadlocks.

2.7.0

neo4j-graph-data-science-2.7.0

Breaking changes

• Removed support for Neo4j DBMS v4.

New features

• A new parameter targetNodes has been introduced to gds.shortestPath.dijkstra procedures that allows specifying multiple targets rather than a single target.
• GDS Arrow to database import now also supports creating database with block storage engine. This requires at least Neo4j 5.18.
• gds.graph.export now supports setting a database format using the dbFormat key. See https://neo4j.com/docs/operations-manual/current/database-internals/store-formats/ for supported formats (block requires at lest Neo4j 5.18).
• Added a dedicated Arrow export command to extract node labels from an in-memory graph.
• Arrow graph export commands support exporting node identifiers mapped to a consecutive id space, i.e. [0..nodeCount).
• Add support for DBMS V5.21.0

Bug fixes

• Fixed a bug where the batch size configured in arrow stream messages would be ignored
• Fixed a bug where graphs could be projected with empty name when using Cypher projections v2.

Improvements

• Writing relationships can now be done in parallel, the concurrency can be controlled by the writeConcurrency parameter.
• Progress logging for Closeness centrality is more detailed.

Other changes

• The record_format field in the GDS Arrow Create Database action is deprecated for removal, use db_format instead.

Graph Data Science 2.6.8

neo4j-graph-data-science-2.6.8

Bug fixes

• Fixed a bug in the longest path algorithm, where it was failing with zero weights for some of the cases.
• Fixed an issue when Louvain would fail on very big graphs.

Graph Data Science 2.6.7

This release brings compatibility with Neo4j 5.20.0.

Graph Data Science 2.6.6

neo4j-graph-data-science-2.6.6

Bug fixes

• Fixed an issue when valid procedure calls wouldn't run when using Cypher Parallel Runtime.
• Fixed an issue where gds.listProgress() would not contain entries for running procedures.

2.6.5

New features

• Added compatibility with Neo4j 5.19.0.

2.6.4

Bug fixes

• Fixed a bug in the Arrow server token cache where authentication could fail sporadically with correct login credentials.
• Fixed a bug in the Arrow server where a wrongly configured encryption key could cause the database server to fail to start.
• Adapted an internal buffer resize strategy to avoid potential overflows when creating graphs with very high degree nodes (e.g. in Louvain).

2.6.3

Bug fixes

• Fixed a bug where gds.graph.project would fail when a default value for an array property was used across multiple labels.
• Fixed a bug where GraphSage memory estimations would fail.
• Fixed a bug where gds.graph.project would sometimes not read all data when the database was using the block storage format.