Skip to content

Release 3.4

Choose a tag to compare

View all tags
@doutriaux1 doutriaux1 released this 24 Jul 21:09
· 2 commits to stable since this release
v3.4
6fb8f54

3.4 Release

Description

This is a minor release which adds parameterized workflow tracking and includes clustering and store robustness fixes.

New in this release

  • Added the kosh_workflow command line tool for creating, querying, and updating datasets from a workflow step plus a parameter set.
  • Added the kosh.parameter_store.StepRequest and kosh.parameter_store.apply_step(...) Python API for the same workflow-tracking behavior in scripts.
  • Workflow matching supports optional ensemble scoping, exact numeric matching with --strict-match, tolerance-based numeric matching with --rtol and --atol, duplicate avoidance with --upsert/--upsert-init, dataset and ensemble metadata updates, and file associations.
  • kosh_workflow --check reports matching dataset ids, current workflow step, and last step-update timestamp; kosh_workflow --size reports store or ensemble dataset counts.
  • Added examples/Example_Workflow_Manager.ipynb and documentation references for parameter-set workflow management.
  • Packaged the Kosh Codex skill files and documented workflow tracking guidance for agent use.

Improvements

  • Kosh now requires Sina 1.16.0 or greater.
  • Modernized package metadata through pyproject.toml and entry points for kosh and kosh_workflow.
  • Improved cleanup on store close to reduce SQLAlchemy/SQLite shutdown warnings when connections are finalized from a different thread.
  • Made Kosh user-record creation safer under concurrent access by treating confirmed duplicate user records as success.
  • Normalized empty MPI clustering index results to an empty array instead of None, which makes downstream gather/concatenate operations more reliable.

Bug fixes

  • Fixed clustering failures caused by single-value NumPy arrays being passed where scalar distance thresholds or loss values were required.
  • Made batch clustering tests compare against a non-batched reference instead of dependency-sensitive hard-coded sample counts.
Assets 2
Loading