Skip to content
Permalink
Branch: master
Commits on Jul 16, 2019
  1. Test `lambda_callback` provision methods

    HunterMcGushion committed Jul 16, 2019
    - Check that different ways of providing `lambda_callback`
      produce identical `CVExperiment` MROs
  2. Add `callbacks` kwarg to `CVExperiment`

    HunterMcGushion committed Jul 16, 2019
    - Enables passing custom callbacks directly on `CVExperiment`
      initialization as an alternative to `Environment`'s
      `experiment_callbacks`, which are applied to all Experiments/OptPros
    - This kwarg is admittedly pretty weird because although it is documented
      in `BaseExperiment`, it only actually exists for `CVExperiment`
    - Furthermore, `CVExperiment` itself doesn't do anything with the
      `callbacks` kwarg
    - It is instead intercepted by `ExperimentMeta` and applied to the
      `CVExperiment`, as the comment next to the new `callbacks` kwarg notes
    - Also updated `Environment`'s `experiment_callbacks` documentation
Commits on Jul 14, 2019
  1. Merge pull request #162 from HunterMcGushion/release/3-0-0-beta-0

    HunterMcGushion committed Jul 14, 2019
    Release v3.0.0beta0
  2. Release v3.0.0beta0

    HunterMcGushion committed Jul 14, 2019
  3. Merge pull request #159 from HunterMcGushion/maint/forge-experiment

    HunterMcGushion committed Jul 14, 2019
    Rename `set_experiment_guidelines` to `forge_experiment`
  4. Update changelog entry for `model_init_params` as optional

    HunterMcGushion committed Jul 14, 2019
    - Updated to include the fact that `model_init_params` is also
      optional for `BaseOptPro.forge_experiment`
  5. Unify docs for `BaseExperiment.__init__` and `BaseOptPro.forge_experi…

    HunterMcGushion committed Jul 13, 2019
    …ment`
    
    - Update both to reflect `forge_experiment` name change
    - Follow same format in both and link to each other
  6. Support `feature_engineer` kwargs as list

    HunterMcGushion committed Jul 13, 2019
    - Allows the `feature_engineer` kwarg of both `CVExperiment` and
      `BaseOptPro.forge_experiment` to be provided as a list of `steps`,
      rather than a `FeatureEngineer` instance
    - Added tests to ensure result is the same for both OptPros and
      Experiments regardless of the type of `feature_engineer`
    - :meth:`FeatureEngineer.__eq__` added to facilitate testing
  7. Add changelog entries for bugfix and `optimization` overhaul

    HunterMcGushion committed Jul 10, 2019
    - Entry for `BayesianOptPro`/exclusively-`Categorical` bug fix
    - Entries for `optimization` and `space` module restructuring
  8. Formally deprecate `BaseOptPro.set_experiment_guidelines`

    HunterMcGushion committed Jul 10, 2019
    - Renamed to `forge_experiment`
    - `set_experiment_guidelines` remains as a deprecated method,
      that defers to `forge_experiment`
    - `set_experiment_guidelines` is scheduled to be removed in 3.2.0
    - Add unit tests for proper deprecation of `set_experiment_guidelines`
  9. Document/refactor `test_protocols`

    HunterMcGushion committed Jul 10, 2019
    - Add documentation for parametrization functions
    - Convert dummy `Environment` fixture to an actually usable one
    - Minor refactoring
  10. Document `base_estimator` validation tests

    HunterMcGushion committed Jul 10, 2019
    - Also rename test functions to be more descriptive
  11. Rename `set_experiment_guidelines` to `forge_experiment`

    HunterMcGushion committed Jul 9, 2019
    - Aside from the obvious benefit of sounding way more badass, this
      change was also made to better clarify some of the library's terminology
    - `set_experiment_guidelines` was particularly problematic because the term
      "guideline" is used to describe hyperparameters whose values are given in
      terms of a constant value, rather than a `Dimension` subclass (which is often
      referred to as a "choice")
    - This is especially prevalent in `utils.optimization_utils.filter_by_guidelines`,
      as well as `result_reader.ResultFinder`, which defines methods
      `_filter_by_guidelines` and `_filter_by_guidelines_multi`
    - This change helps cement "guidelines" and "choices" as the two methods of providing
      hyperparameters to an OptimizationProtocol, by removing ambiguous uses of the
      "guidelines" term
Commits on Jul 11, 2019
  1. Re-update ReadTheDocs badge

    HunterMcGushion committed Jul 11, 2019
  2. Update ReadTheDocs/Coveralls badges

    HunterMcGushion committed Jul 11, 2019
    - ReadTheDocs badge now links to "stable" build, instead of "latest"
    - Attempt to fix Coveralls badge stuck at old values
Commits on Jul 9, 2019
  1. Add `EngineerStep` comparison with str

    HunterMcGushion committed Jul 9, 2019
    - Allow equality testing between `EngineerStep` and
      strings of form produced by :meth:`EngineerStep.stringify`
    - Also refactor `EngineerStep.__eq__` to utilize
      `EngineerStep.get_comparison_attrs` for comparisons against
      other `EngineerStep` instances or dicts
  2. Fix references to old `optimization_core`

    HunterMcGushion committed Jul 9, 2019
  3. Merge pull request #156 from HunterMcGushion/bug/all-cat-bayesian

    HunterMcGushion committed Jul 9, 2019
    Fix all-`Categorical`/`BayesianOptPro` bug
  4. Add test for `Environment.__repr__`

    HunterMcGushion committed Jul 9, 2019
  5. Remove `_optimization_loop` StopIteration catching

    HunterMcGushion committed Jul 9, 2019
    - This is a remnant from `UninformedOptimizationProtocol`, whose
      :meth:`_get_current_hyperparameters` would try to return
      `next(self.space)`, which would, in turn, raise `StopIteration`
      once `space` had been exhausted
    - This is no longer necessary since `UninformedOptimizationProtocol`
      was never finished and removed long ago
    - If reimplementing a standard grid search protocol (which sounds
      like a pretty good idea), refer to the content removed in this
      commit, as well as commits 9b7ca73 and e2c3b73 from October 25, 2018
      for some "pro" tips
Commits on Jul 8, 2019
  1. Make xfail tests for bug in #154 conditional

    HunterMcGushion committed Jul 8, 2019
    - These tests are now only expected to fail if the HyperparameterHunter
      version is <= "3.0.0alpha2"
    - These tests will still be marked as `xpassed`, rather than simply
      `passed` until the VERSION file has been bumped
  2. Fix `BayesianOptPro`/all-`Categorical` bug. Closes #154

    HunterMcGushion committed Jul 8, 2019
    - This commit makes two small changes, which, when taken with
      the optimization/space restructuring of #153, resolves #154
    - For posterity, the two changes made herein were tested prior
      to #153 to no avail
    - Suggests that there may be some aspect of the refactoring of
      SKOpt code in #153, which helped to resolve the underlying bug
Older
You can’t perform that action at this time.