0.1.8a0

What's Changed

• sync to main by @perib in #141
• Adding SimpleImputer, IterativeImputer and KNNImputer to the config space. by @gketronDS in #142
• allow genetic feature selection to work with nan values by @perib in #143
• Update docs.yml by @nickotto in #144
• Dev by @perib in #145
• added fixed bins, fixed bugs by @perib in #147
• Dev by @perib in #148
• Update _version.py by @perib in #149

New Contributors

• @gketronDS made their first contribution in #142

Full Changelog: 0.1.7a0...0.1.8a0

0.1.7a0

What's Changed

• New search space def by @perib in #122
• New search space def by @perib in #123
• fixes, passing tests by @perib in #124
• more fixes with search spaces - wrapper, make sure all supported modu… by @perib in #126
• added unique_id, fixed issue where graphpipeline not correctly identi… by @perib in #127
• New search space def by @perib in #128
• New search space def by @perib in #129
• fix num_nodes test, add update checker, setup.py change by @perib in #130
• New search space def by @perib in #131
• New search space def by @perib in #132
• remove hardcoded wrappers that are no longer needed by @perib in #133
• New search space def by @perib in #134
• New search space def by @perib in #136
• New search space def by @perib in #137
• New search space def by @perib in #138
• New search space def by @perib in #139
• Search space api by @perib in #140

Full Changelog: 0.1.6a1...0.1.7a0

0.1.6a1

What's Changed

• setup.py was updated with new requirement versions for numpy, dask. Python version < 3.12 is required to ensure compatibility with configspace

Full Changelog: 0.1.6-alpha...0.1.6a1

0.1.6-alpha

What's Changed

• Fixed Tutorial 8
• Updated to pandas 2.2.0+
• Eval error moved to a new column
• Fixed the parent selection function (steady state)

• sync dev by @perib in #116
• Merge pull request #116 from EpistasisLab/main by @jay-m-dev in #118
• fix turorial 8 by @perib in #119
• Pandas update by @nickotto in #121
• Dev by @perib in #120
• Dev by @jay-m-dev in #125

Full Changelog: 0.1.5-alpha...0.1.6-alpha

0.1.5-alpha

What's Changed

• fixed bug preventing recursive search spaces - by @perib in #113
• fixed config dict typo preventing lists of configuration dictionaries from being used - by @perib in #114

Full Changelog: 0.1.4-alpha...0.1.5-alpha

v0.1.4-alpha release

What's Changed

• Hyperparameter mutation probabilities and gradual changes by @perib in #103
• selection fix when zero mutation/cx selected by @perib in #105
• Increase reproducibility and control stochasticity by @jgh9094 in #106
• fix to allow mutations to actually occur to an individual by @jgh9094 in #108
• Random initialization by @jgh9094 in #107
• bug fixes by @perib in #109
• preprocessing fix by @perib in #110

Full Changelog: 0.1.3-alpha...0.1.4-alpha

v0.1.3-alpha release

What's Changed

• API changes on the Base Evolver
• SteadyStateEvolver integration and updates
• New Tutorials and updates
• Updated the base evolver for improved functionality.
• Enhanced the Dask code for increased stability.
• Added the sklearnex library and support.

Changes to BaseEvolver:

• Sample probabilities for mutation and crossover individually, rather than using fixed frequencies.
• Reorganized step() and one_generation_step(), moving survival selection and evaluation to step() and adding generate_offspring() for more flexibility in creating offspring.
• Introduced generate_offspring() to address customization of offspring creation, simplifying modifications.
• Integrated survival selection and parent selection in the population class to streamline user interaction at the evolver level.
• Enabled the use of individual mutation and crossover functions and their respective weights.
• Maintained support for both "mutate then crossover" and "crossover then mutate" methods, with consideration for potential code complexity.

Other Changes:

• Updated the parallel eval function for the base evolver to handle futures and timeouts individually.
• Reduced the fallback timeout to 1.25 times max_eval_time_seconds.
• Resolved issues with Jupyter notebooks for GitHub by removing problematic markdown syntax.
• Noted that the steady-state code remains unchanged, with a suggestion to consider altering the population class for improved clarity and functionality by separating evaluated parents from unevaluated offspring. Currently, both are stored in the same list within the population class.

Full Changelog: 0.1.2-alpha...0.1.3-alpha

v0.1.2-alpha

API changes:
Removed max_children and max_depth.
Added linear_pipeline (bool). If True, it will evolve linear pipelines.
Changed inner_config_dict: If set to None, your graph should have no inner nodes. (Previously, if None, inner nodes would be pulled from root_config_dict)

Bug fixes:

In rare circumstances, some pipelines could not be timed-out correctly with max_eval_time_seconds. Additionally, the overall timeout from max_time_seconds was only checked at the end of a generation. Both of these could lead to long runtimes. Now TPOT2 can terminate the evaluation within a generation so that the final runtime is closer to max_time_seconds. There is also a fallback timeout to terminate the generation if it appears that the threads are not being terminated in time.

Optimized calculation of the pareto frontier.

Some other minor internal edits/cleanups/fixes.

Known Issues:

The fallback timeout is currently not compatible with the population evaluation progress bar (though the generation progress bar works correctly). Temporarily set the evaluation progress bar to be verbose>=6. If the progress bar is used, the fallback timeout will not work.

pypi init

0.1.1-alpha

Update _version.py

init

Merge pull request #6 from perib/dev

allow survival selector to be none