Skip to content

CHANGELOG.md

Latest commit

 

History

History
82 lines (73 loc) · 3.04 KB

CHANGELOG.md

File metadata and controls

82 lines (73 loc) · 3.04 KB

Changelog

All notable changes to Emukit will be documented in this file.

[0.5.1]

  • Use scipy for model free designs (#483)
  • Fix division by 0 warning (#481)
  • Document release process (#479)

[0.5.0]

  • Upgrade codebase to allow numpy versions greater than 2.0, while maintaining backwards compatibility. GPy remains pinned to numpy<2.0 due to upstream constraints, but core Emukit functionality is now independent of it. Users can now install and use Emukit core API with numpy>=2.0.
  • Packaging: Adopt PEP 621 metadata in pyproject.toml; dynamic version from emukit.__version__.
  • Packaging: Introduced setuptools extras (gpy, bnn, sklearn, docs, examples, tests, dev).
  • CI: Workflows now install extras via pip install -e .[tests] (and [tests,gpy]) instead of requirements files.
  • Docs: Updated installation guide, README, CONTRIBUTING to prefer extras over legacy requirements/ files.
  • Tests: Documented pytest marker and optional dependency usage.
  • Docs build: Use docs extra (includes GPy).
  • Maintenance: Legacy requirements files retained temporarily for reference.

[0.4.11]

  • Various bugfixes, including installation on Windows
  • Updated copyright info

[0.4.10]

  • Wrapper for SKlearn Guassian process
  • Black and isort formatting
  • Brownian motion quadrature kernel and product embedding
  • ProductMatern52 quadrature kernel embedding
  • Multiple improvements to quadrature integration measures
  • QuadratureProductKernel base class
  • Doc improvements
  • Bug fixes, including scipy compatibility fixes

[0.4.9]

  • Update to newest version of GPy, which shall fix installation issues
  • Mean Plug-in Expected Improvement
  • Square root warping for BQ and WSABI
  • Improved validation of categorical variables
  • Updates and fixes of Local Penalization acquisition function
  • bug fixes
  • doc fixes

[0.4.8]

  • Added sobol initial design
  • BanditParameter
  • Boolean operations for stopping conditions
  • Preferential Bayesian optimization example
  • MUMBO acquisition function
  • Revised dependecies' versions requirements
  • Bug fixes
  • Doc fixes

[0.4.7]

  • Added simple GP model for examples
  • Bayesian optimization with unknown constraints
  • Removed dependency on libomp
  • Max value entropy search acquisition function
  • Multi point expected improvement acquisition function
  • Moved model free designs to core
  • Profet implementation
  • Added citation info
  • QRBF for uniform and Gaussian measures
  • uncertainty sampling acquisition for bq
  • Bayesian Monte Carlo
  • Bugfixes
  • Doc fixes

[0.4.6]

  • Added support for inequality constraints
  • Fabolas as an example
  • Bugfixes

[0.4.5]

  • Confirmed support for Python 3.7
  • Removed dependency on GPyOpt
  • Implemented generic IntegratedHyperParameterAcquisition
  • Added notebooks validation automation
  • Random baseline for benchmarking
  • Implemented a range of discrete optimizers
  • Uniform measure and mutual information acquisition for BQ
  • Added sample_uniform method to parameter space and individual parameter types
  • Improved unit test coverage
  • Various fixes in code, comments and notebooks