Skip to content

v2.0

Latest
Latest

Choose a tag to compare

View all tags
@FlorentinD FlorentinD released this 18 Sep 14:10
· 1 commit to 2.0 since this release
2.0
cf282c5
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Changes in 2.0

Breaking changes

  • All untyped 1.x endpoints have been removed. The V2 API, previously available under the gds.v2 prefix, is now the default and only API of the client, and the gds.v2 prefix is gone. See the migration guide for how to port existing code: https://neo4j.com/docs/graph-data-science-client/current/migration-from-1x/
  • The minimum supported GDS server version is now 2.13.0.
  • The minimum supported Neo4j Python driver version is now 5.26.0. The Neo4j 4.4 driver is no longer supported.
  • Pandas 1.x is no longer supported. The supported range is now pandas >= 2.0, < 4.0, which includes Pandas 3.0.
  • The supported pyarrow range is now >= 21.0, < 26.0, and the supported numpy range is now < 3.0.
  • The failIfMissing parameter of Graph.drop and Model.drop was renamed to fail_if_missing.
  • ServerVersion and SemanticVersion have moved to the graphdatascience.versions package.
  • The write_concurrency and job_id parameters have been removed from gds.graph.node_labels.mutate.
  • ArrowEndpointVersion.from_arrow_info has been replaced by check_version_compatibility.
  • The retryable parameter of run_cypher has been removed. Cypher queries run via run_cypher now always use transactional retries.
  • GraphV2 was renamed to Graph and ModelV2 was renamed to Model. Graph is now exported at the top level, so from graphdatascience import Graph works.
  • For Aura Graph Analytics gds.graph.project was renamed to gds.graph.project.cypher
  • For Aura Graph Analytics gds.graph.project_native was renamed to gds.graph.project.native
  • Provide reason for deleted state of a deleted GDS Sessions
  • The database parameter has been removed from gds.graph.project.cypher and gds.graph.project.cypher_async. The database is now always sourced from the gds object; use gds.set_database() to change it.
  • gds.graph.drop now returns a list of GraphInfo objects as it now allows to delete multiple graphs at once.

New features

  • Added an algorithms parameter to GdsSessions.estimate to estimate the memory for individual algorithms and their configuration, instead of whole algorithm categories. The parameter accepts either a list of algorithm names or a mapping from algorithm name to configuration. Names are matched case-insensitively, the gds. prefix is optional, and Python endpoint names such as node_embedding.fastrp are mapped to their GDS procedure names.
  • Added gds.pipeline.get to retrieve a pipeline from the pipeline catalog, and pipeline objects can now return their pipeline info.
  • Added support for the HITS algorithm (gds.hits) in Aura Graph Analytics sessions, with stream, stats, mutate and write modes.
  • Added support for the topological link prediction functions (gds.topological_link_prediction) in Aura Graph Analytics sessions.
  • GdsSessions.get_or_create now accepts a show_progress parameter to control whether the returned client prints its own progress bars.
  • Added an optional overwrite parameter to gds.graph.project, gds.graph.generate, gds.graph.construct, gds.graph.filter and gds.graph.sample (and their async/session variants). When set to True, an existing graph with the same name is dropped before the new graph is created.
  • Expose gds.fast_path for Aura Graph Analytics / GDS Sessions.

Bug fixes

  • Requests to the Aura API that are rejected as unauthorized are now retried once with a new OAuth token, instead of failing permanently while the expired token stays cached
  • Requests to the Aura API now use a connect and read timeout, so that a stalled connection can no longer block a call indefinitely.
  • GdsSessions.delete now returns False if no session was deleted when called with a session_id. It previously always returned True.
  • AuraApiError and SessionStatusError no longer include a repetition of the exception object in their message.
  • The Arrow endpoint version is now checked when the client is created. If it is unsupported, the client raises an error asking to update the graphdatascience package, instead of failing later with an unrelated error.
  • Getting a session that has already expired now raises a RuntimeError with a human-readable duration, instead of warning with a wrapped (and misleading) timedelta.seconds value. Sessions expiring within the next hour still emit a warning.
  • GraphDataScience.close() now closes the Arrow Flight client in addition to the query runner.
  • gds.graph.construct now correctly forwards the inverse_indexed_relationship_types.

Improvements

  • GraphDataScience no longer requires the aura_ds parameter to be set. If left unset, the client automatically derives whether the database is hosted in Aura.
  • gds.project.cypher will automatically rewrite queries that contain gds.graph.project instead of gds.graph.project.remote
  • gds.project.cypher will check if undirectedRelationshipTypes and inverseIndexedRelationshipTypes are defined in the projection query instead of the method parameters.
  • The show_progress setting of a client is now honoured consistently: graph projections, WriteJobHandle.wait and the node property endpoints all inherit it. ProgressBar.set_default_options allows setting process-wide progress bar options.
  • Estimation errors and documentation now refer to Python endpoint names instead of GDS procedure names.
  • Session errors, such as a session failing with an out-of-memory error, are now reported together with the session status. They are also surfaced automatically when an operation fails because the session can no longer be reached, instead of only reporting the underlying connection error.
  • Improve error message if gds.graph.project.cypher produced no graph due to no matching data.

Other changes

Assets 2
Loading