Propulate v1.2.2

What's Changed

New Features

• Enable support for internally parallelized evaluation of the loss function using multi-rank workers by @mcw92 in #110
• Provide example script for how to use Propulate with multi-rank workers for HPO / NAS of distributed data-parallel neural networks in PyTorch by @coquelin77 in #124
• Introduce CMA-ES propagator by @DonnyDevV in #91
• Introduce particle swarm optimization propagator by @Morridin in #78
• Introduce Nelder-Mead propagator by @oskar-taubert in #125
• Introduce surrogate models for early stopping of unpromising individuals by @vtotiv in #112
• Provide improved example script for how to use Propulate for HPO / NAS of neural networks in PyTorch (see torch_example.py) by @mcw92 in #106

Maintenance

• Introduce integration tests by @oskar-taubert in #98 and #99 and #116 and #140 and @mcw92 in #122
• Enable functional pre-commit hooks by @mcw92 in #114
• Improve tutorials and documentation by @mcw92 in #138

New Contributors

• @DonnyDevV made their first contribution in #91
• @Morridin made their first contribution in #78
• @vtotiv made their first contribution in #112

Full Changelog: v1.0.1...v1.2.2

v1.1.0

What's Changed

New Features

• Provide automatically built Sphinx documentation including installation instructions, theoretical background, tutorials, and API references at https://propulate.readthedocs.io/ by @mcw92 in #71
• Add comprehensive tutorials
  • Add example script for how to use Propulate for HPO / NAS of neural networks in PyTorch by @oskar-taubert in #75
  • Iss12 - Add simple example script without islands by @mcw92 in #54
  • Include benchmark functions from Propulate publication by @mcw92 in #69
• Introduce separate logging for Propulate optimizer, enabling using a separate logger within the loss function for, e.g., NAS by @coquelin77 in #73
• Introduce clean and more intuitive checkpointing
  • Iss44 - Create checkpoint path in Pollinator if not exists by @mcw92 in #51
• Add contribution guidelines by @mcw92 in #79
• Add typehints and clean and consistent docstrings by @mcw92 in #64
• Iss40 - Boundary-inclusive sampling for ordinal parameters by @mcw92 in #52
• Iss46 - Introduce evaluation time as Individual attribute #46 by @SMEISEN in #47

Maintenance

• More cleaned-up, structured, and refactored code, including consistent docstrings, type hints, and meaningful names for classes and variables
  • Introduce Propulator base class by @mcw92 in #74
  • Iss39 - Remove inter-island communicator by @mcw92 in #53

New Contributors

• @SMEISEN made their first contribution in #47

Full Changelog: 1.0.1...v1.1.0

v1.0.1

What's Changed

• Create first release by @oskar-taubert in #43 and #42

Full Changelog: 1.0.0...1.0.1

v1.0.0

What's Changed

• Provide new lightning example for HPO / NAS of neural networks in PyTorch by @oskar-taubert in #25
• Rename SelectBest/SelectWorst to SelectMin/SelectMax by @mcw92 in #27
• Rename wrapper to islands by @mcw92 in #29
• Create user-provided checkpoint folder if not exists by @mcw92 in #33

Bug fixes

• Fix rng.sample() call for Python 3.11 by @elcorto in #15
• Iss34 - Fix default propagator for ordinal-only HPO by @mcw92 in #37
• Iss38 - Fix continuing from checkpoint by @mcw92 in #41
• Fix checkpoint path by @oskar-taubert in #26

New Contributors

• @elcorto made their first contribution in #15

Full Changelog: https://github.com/Helmholtz-AI-Energy/propulate/commits/1.0.0