Skip to content

What's New?

1.12.1 (2024-08-13)#

Fix#

  • pipeline: request now fails without default (#284)

1.12.0 (2024-04-24)#

Feat#

  • PyTorch: Add functionality to construct a PyTorch Model from a pipeline (#276)

Fix#

  • Pass in sampler to create_study (#282)
  • Pytorch: Fix builders.py (#280)
  • precommit issues from #276 (#277)

1.11.0 (2024-02-29)#

Feat#

  • CVEvaluator: Add feature for post_split and post_processing (#260)
  • sklearn: X_test, y_test to CVEvaluator (#258)
  • CVEarlyStopping (#254)
  • sklearn: CVEvaluator allows configure and build params (#250)
  • sklearn: Provide a standard CVEvaluator (#244)

Fix#

  • trial: Don't record metric values for deserialized NaN's or None (#263)
  • pipeline: Ensure optimizer is updated with report (#261)
  • scheduling: Safe termination of processes, avoiding lifetime race condition (#256)
  • metalearning: Portfolio Check for Dataframe as Input (#253)
  • CVEvaluator: clone the estimator before use (#249)
  • Node: Ensure that parent name does not conflict with children (#248)
  • CVEvaluator: When on_error="raise", inform of which trial failed (#247)
  • Trial: Give trials a created_at stamp (#246)

Refactor#

  • pipeline: optimize now requires one of timeout or (#252)
  • Metric, Trial: Cleanup of metrics and Trial (#242)

1.10.1 (2024-01-28)#

Fix#

  • dask-jobqueue: Make sure to close client

Refactor#

  • Make things more context manager
  • trial: Remove begin() (#238)

1.10.0 (2024-01-26)#

Feat#

  • Pipeline: Optimize pipelines directly with optimize() (#230)

1.9.0 (2024-01-26)#

Feat#

  • Optimizer: Allow for batch ask requests (#224)

Fix#

  • Pynisher: Ensure system supports limit (#223)

1.8.0 (2024-01-22)#

Feat#

  • Pynisher: Detect tasks with Trial to report FAIL (#220)
  • Pipeline: factorize() a pipeline into its possibilities (#217)

1.7.0 (2024-01-16)#

Feat#

  • Scheduler: Respond to cancelled futures (#214)
  • scheduler: Handled errors with specific method (#213)

Fix#

  • History: Explicitly check type in add() (#210)

1.6.0 (2024-01-10)#

Feat#

  • history: Get best() from History (#209)

1.5.0 (2024-01-09)#

Feat#

  • add EmissionsTrackerPlugin for codecarbon (#196)

1.4.0 (2023-12-12)#

Feat#

  • Scheduler: Monitor to view efficiency (#197)

Fix#

  • data: reduce_int_span with nullable dtypes (#200)

Refactor#

  • Scheduling: limit -> max_calls (#201)

1.3.4 (2023-12-07)#

Fix#

  • pipeline: configure only operates on chosen choice (#195)

1.3.3 (2023-12-06)#

Fix#

  • rich: Move import into scoped block (#193)

1.3.2 (2023-12-06)#

Fix#

  • _doc: Catch missing else statement for doc_print (#190)

1.3.1 (2023-12-05)#

Fix#

  • pyproject: Change to recognized classifier for PyPI (#189)

1.3.0 (2023-12-05)#

Feat#

  • History: Default to normalizing time of history output

Fix#

  • Provide more information if built_item fails (#187)
  • _doc: Use isinstance on types
  • Optimizers: Default to optimizer name #174

Refactor#

  • History: History provides mutator methods
  • Move StoredValue to own file in .store

1.2.4 (2023-11-25)#

Fix#

  • docs: Optimizer inline examples (#172)

1.2.3 (2023-11-24)#

Fix#

  • Trial: Add table to rich renderables (#170)
  • dask-jobqueue: Default to scale() for predictable behaviour (#168)

1.2.2 (2023-11-23)#

Fix#

  • test: remove stray output from test/docs

1.2.1 (2023-11-22)#

Fix#

  • pipeline: request() correctly sets config key
  • Scheduler: Make sure it's displayable in notebook

1.2.0 (2023-11-20)#

Feat#

  • sklearn: Special keywords for splits

Fix#

  • trials: Always use a PathBucket with optimizers
  • Trial: Trial should now always have a bucket

1.1.1 (2023-11-19)#

Fix#

  • doc: Add classifiers to pypi for bades (#155)

1.1.0 (2023-11-19)#

Feat#

  • scheduler: method call_later (#145)

Refactor#

  • optimization: Add concept of Metric (#150)

1.0.1 (2023-11-15)#

Fix#

  • CHANGELOG: Fresh changelog