Graph Data Science 2.15.0

New features

• gds.articulationPoints.stream can now return for each articulation point a map resultingComponents tracking the number of components, as well as their minimum/maximum size that are formed following the articulation point's removal.

Bug fixes

• Fixed a bug where algorithms would keep tasks around on failure. Affected algorithms:
  • gds.beta.pipeline.linkPrediction.train - Could not start another training on the same graph afterwards.
  • gds.dag.longestPath
  • gds.randomWalk
  • gds.shortestPath.yens
  • gds.allShortestPaths.delta

Improvements

• Native projections will now reject relationships which point to or from nodes that were not scanned during the node projection.
  This situation can happen when there are simultaneous modifications to the DBMS during projection.
• In gds.fastRP.{mutate,write,stream}, the initial random node vectors use a degree-scaling which is uniform over the entire vector, whereas previously a different scaling was applied to the property-aware part of the vector.
  This makes the relationship between propertyRatio and the influence of properties more intuitive, also when normalizationStrength is non-zero.