v0.3.1

This release mainly introduces improvements under the hood, including better typing and other optimizations. Expensive operations such as the computation of lower and upper entropy can now be parallelized. Additionally, probly now uses pytraverse to traverse neural networks provided by the user, which allows it to parse almost any custom neural network.

What's Changed

• Improve the docs by @pwhofman in #102
• Parallelize the lower, upper entropy functions, improve docstrings, c… by @pwhofman in #108
• optimize workflows by @mmschlk in #109
• Improve docs, add tutorials, and improve docstrings by @pwhofman in #110
• Integration of the Traverser API into the representation module by @Cortys in #112

Full Changelog: v0.3.0...v0.3.1

v0.3.0

This release introduces new uncertainty representations based on credal sets, support for credal sets based on probability intervals and the convex hull, support for datasets with probability distributions as targets, many calibration methods, and more.

What's Changed

• This PR should add Docs to the Project by @mmschlk in #65
• Forces from __future__ import annotations import and fixes type hints import by @mmschlk in #66
• Add metrics based on proper scoring rules that can also be used for u… by @pwhofman in #67
• 34 add temperature scaling for calibration by @pwhofman in #69
• Add Label Relaxation loss, closes #31 and closes #8 by @pwhofman in #70
• Clean up pyproject.toml file by @pwhofman in #72
• 73 add support for credal sets based on the convex hull by @pwhofman in #74
• 49 add sampling to the evidential regression representation by @pwhofman in #75
• adds a new github workflow checking if the install works by @mmschlk in #76
• 38 add data loaders for common first order datasets by @pwhofman in #77
• Add CONTRIBUTING.md and link in the README by @pwhofman in #78
• Tests by @pwhofman in #81
• 82 add mypy static type checker by @pwhofman in #83
• Add first iteration of an uncertainty representation by @pwhofman in #80
• More tests by @pwhofman in #85
• Change efficiency to be maximimal for a point prediction and minimal … by @pwhofman in #88
• Credal utils by @pwhofman in #97
• Fix lower entropy bug, add tests to check for the simplex vertices case by @pwhofman in #91
• changed ruff format selector to ALL by @mmschlk in #96
• 98 add credal ensembling representation by @pwhofman in #100

Full Changelog: v0.2.0-pre-alpha...v0.3.0

Initial release v0.2.0

This is the first release of probly! We introduce multiple common uncertainty representation and quantification methods for machine learning. In this initial release, the uncertainty representations are tailored towards use with the PyTorch library, but this will change in the future. The other features, such as the many quantification methods, rely on NumPy. More to follow soon!

What's Changed

• Add regression example notebook to probly by @timoverse in #39
• add-type-hints, closes #40 by @timoverse in #42
• 24-add-dropconnect-representation-method by @timoverse in #46
• 41-streamline-subensemble-behavior-with-ensemble-representation by @pwhofman in #44
• 33-add-focal-loss by @timoverse in #50
• Adds test workflows and applies code-quality. by @mmschlk in #55
• Add simple conformal prediction classification representation and cov… by @pwhofman in #53
• Adds Building Releases to CI by @mmschlk in #56
• 21 add bayesian neural networks by @pwhofman in #51
• Added support for devices other than cpu for Bayesian representation by @pwhofman in #59
• Add the possibility to initialize the prior using the weights of a tr… by @pwhofman in #62
• enforce-good-code-style by @timoverse in #63
• 47 implement functions for general prediction and uncertainty aware predictions by @pwhofman in #64

New Contributors

• @timoverse made their first contribution in #39
• @pwhofman made their first contribution in #44
• @mmschlk made their first contribution in #55

Full Changelog: https://github.com/pwhofman/probly/commits/v0.2.0-pre-alpha