v0.1.9

What's Changed

• fix: numpy.random.RandomState is no longer an accepted random state b… by @thededlier in #50
• update setup configs by @thededlier in #51
• Update build configs by @thededlier in #52

Full Changelog: v0.1.8...v0.1.9

Auto-Surprise v0.1.8

tl;dr Updated dependencies and cleaned up required libraries

What's Changed

• Bump urllib3 from 1.25.9 to 1.26.5 by @dependabot in #41
• Bump babel from 2.8.0 to 2.9.1 by @dependabot in #42
• Create workflow for tests by @thededlier in #44
• Cleanup requirements and create v2 config for docs by @thededlier in #45
• Bump version by @thededlier in #46

Full Changelog: 0.1.7...v0.1.8

Auto-Surprise v0.1.7

• Rename continuous_parallel strategy to it's correct name - SMBO
• Early stop if baseline loss not reached after a set number of iterations
• Bot security version updates

Auto-Surprise v0.1.6

• Can now set random seed to reproduce experiment results
• Updated requirements to have rich
• Updated documentation to include benchmark results

Auto-Surprise v0.1.5

• Conditionally define similarity options space. This allows for more precise hyper parameter tuning as the number of parameters are reduced in some configurations.
• Pretty print results table
• Attach trials to the final result. This could be used in the future to allow for early stopping.

Auto-Surprise v0.1.4 - Minor Feature Update

• Ability for user to select which algorithms to be optimized with the algorithms argument of Engine
• Replaced debug option of Engine with verbose. This also controls verbosity for Surprise and Hyperopt

Auto-Surprise v0.1.3 - Bug Fixes

v0.1.2 had a breaking bug that swapped the values given for time_limit with max_evals. If you are using v0.1.2 please update to this version. Sorry for any inconvenience.

Auto-Surprise v0.1.2 - Minor Updates

• Updated logging: better logging of results of each iteration for each algorithm.

Auto-Surprise v0.1.1

• BaselineOnly now also has baseline options for hyperparameter optimisation
• Default cross validation iterations changed to 5
• Fix results for non parameterised algorithm results not logged

Initial Release

Initial release which allows for automated algorithm selection and hyper-parameter tuning.