Liesel v0.2.9

• 🚧 Pinned scipy version to <=1.12 (@jobrachem)
• 🚧 Handeled error in liesel.model.model.transform for duplicated nodes (#167, @GianmarcoCallegher)
• ✨ lsl.Calc nodes will now try to evaluate their function upon initialization. This can make it much easier to spot errors early. The behavior can be turned off by passing update_on_init=False. (#92, @jobrachem)

liesel v0.2.8

• ✨ Added NamedTupleInterface (#151, @wiep)

liesel v0.2.7

• ✨ Large documentation updates! (#149, #146, #145, #143, #135, #134, @jobrachem; #144, #140, @GianmarcoCallegher)
• ✨ The method gs.Summary.error_df() is now publicly available. (#137, @jobrachem)
• ✨ The class liesel.goose.engine.SamplingResults is now exported via liesel.goose, which means it can be used as gs.SamplingResults. (@jobrachem)
• ✨ Improved the efficiency of the liesel.distributions.mvn_degen.MultivariateNormalDegenerate.from_penalty constructor (#101, @GianmarcoCallegher)
• ✨ Added alternative constructor .from_penalty_smooth to MultivariateNormalDegenerate (#133, @GianmarcoCallegher)
• 🚧 Added observed=True to a pd.DataFrame.groupby() call in goose/summary_m.py to silence a warning due to a deprecation in pandas v2.1.0
• 🚧 Renamed lsl.Param to lsl.param and lsl.Obs to lsl.obs to reflect the fact that those are functions, not classes. The old names are deprecated and scheduled for removal in v0.4.0. (#130, @jobrachem)
• 🚧 Renamed/moved the following symbols. The old names are deprecated and scheduled for removal in v0.4.0. (#136, @jobrachem)
  • liesel.model.goose.GooseModel -> (renamed/moved) liesel.goose.interface.LieselInterface
  • liesel.goose.models.DictModel -> (renamed/moved) liesel.goose.interface.DictInterface
  • liesel.goose.models.DataClassModel -> (renamed/moved) liesel.goose.interface.DataclassInterface
• 🚧 Some maintenance updates to stay up to date with dependencies (#147, #141, #102, @GianmarcoCallegher; #139, @wiep; #109, @jobrachem)

Edit: The version number was too high in __version__.py. I updated the release title to fit the version number. Now there is just no liesel v0.2.6 on pypi.

liesel v0.2.5

• 🚧 Updated for compatibility with Blackjax 1.0.0 (#100, @wiep & @hriebl)
• 🚧 Updated for compatibility with the latest mypy update (#97, @wiep & @hriebl)
• ✨ Added functionality for easy setup and customization of initial value jittering (#72, @GianmarcoCallegher & @hriebl)
• ✨ Improved error messages in lsl.Calc.update() (#84, @jobrachem)
• 🚧 Fixed a bug in gs.plot_param() (#81, @viktoriussuwandi)
• 🚧 Fixed an error in the tutorial on linear regression (#85, @jobrachem)
• 🚧 Fixed the display of the plot title in gs.plot_scatter() (#98, @hriebl)

All commits since v0.2.4

liesel v0.2.4

What's new?

• 🚧 Removed all references to jax.numpy.DeviceArray to make Liesel compatible with Jax 0.4.14 (#73, @jobrachem)
• ✨ Added a visual distinction for edges that represent a connection to a variable's distribution or value (#76, @GianmarcoCallegher)
• ✨ Added ls.Model.simulate(), which provides a convenient way to draw random samples from a Liesel model using the specified priors. (#70, @hriebl)
• ✨ Added liesel.model.goose.finite_discrete_gibbs_kernel, which helps you to automatically set up a gs.GibbsKernel for a discrete variable (#64 & #65, @jobrachem and @hriebl)
• ✨ Added an intialization message to gs.Engine (#66, @GianmarcoCallegher)

All commits since v0.2.3

liesel v0.2.3

What's new?

• 📖 Improved documentation (#47, #48, @jobrachem, @hriebl)
• ✨ Added convert_dtype to graph builder (#50, @hriebl)
• ✨ New overview page for tutorials (#62, @jobrachem)
• ✨ New tutorial on advanced group usage (#63, @jobrachem)
• ✨ Added a method to convert sampling results to arviz's inference data (#49, @wiep)
• 🚧 Changes __repr__ for multiple classes in liesel.model (#57, @wiep)

All commits since v0.2.2

liesel 0.2.2

What's new?

• 🚚 The tutorials have been updated to v.0.2.2 and are now part of the documentation (@jobrachem, @wiep, @hriebl, @GianmarcoCallegher)
• ✨ Added new tutorial showcasing the interface to PyMC (@GianmarcoCallegher)
• ✨ Added node / variable groups (#28, @jobrachem)
• ✨ Sampling from the MultivariateNormalDegenerate is now possible (#34, @jobrachem)
• 🚧 Fixed undefined behaviour in the distreg module (#20, @hriebl)
• 🚧 The distreg module will now use variables names as position keys (#22, @hriebl)

All commits since 0.2.1

liesel 0.2.1

What's new?

• 🦢 Updated the experimental PyMC interface for Goose to PyMC 5.0 (@wiep)
• 🦢 Fixed a bug with the column dtype being object instead of float in the MCMC summary (@wiep)

All commits since 0.2.0

liesel 0.2.0

For this release, the Liesel modeling library has been rewritten from scratch. We are currently working on updated tutorials explaining the new concepts introduced in v0.2.0 in full detail.

What's new?

• 👧 Rewrote the Liesel modeling library from scratch, introducing the Var and the GraphBuilder
• 👧 Replaced the SmoothPrior TFP distribution with the more general MultivariateNormalDegenerate
• 🦢 Removed deprecated functionality from the Goose summary modules
• ⚠️ The import paths have changed:
  • import liesel.liesel as lsl ➡️ import liesel.model as lsl
  • import liesel.tfp.jax.distributions as lsld ➡️ import liesel.distributions as lsld
  • import liesel.tfp.jax.bijectors as lslb ➡️ import liesel.bijectors as lslb

Contributors

• @hriebl
• @wiep
• @jobrachem
• @GianmarcoCallegher

All commits since 0.1.4

liesel 0.1.4

What's new?

• 🌍 We have a new project homepage: https://liesel-project.org
• 📖 Migrated the docs from pdoc to Sphinx: https://docs.liesel-project.org (@jobrachem)
• 🦢 An MCMC summary can now be created with gs.Summary(results) instead of gs.Summary.from_results(results) (#13, @jobrachem)
• 🦢 Fixed a bug with the reported quantiles in the per-chain MCMC summary (#14, @jobrachem)

All commits since 0.1.3