v1.0.0

This 1.0 ArviZ release comes with many breaking changes: See the migration guide for an overview of improvements and guidance on things that will break and how to update them.

The whole library has been refactored for extra flexibility and modularity as well as reducing the cost of maintaining and adding new features to the library. arviz now exposes all the common functions which are implemented in independent packages of the ArviZverse. Take a look at the online book on Exploratory Analysis of Bayesian Models to see it in action and at the docs of the 3 ArviZ packages: arviz-base, arviz-stats and arviz-plots (note all the objects at the top level namespace of these 3 libraries are also exposed as arviz.xyz)

v0.23.4

Extra windows specific fix for once a day warning.

v0.23.3

Path release with improvements to the once a day warning at import time

v0.23.1

Patch release to handle the recent 1.8 h5netcdf release. h5netcdf now has two potential backends, and one must be chosen explicitly to get an install capable of reading/writing netcdf files.

First ArviZ 1.0 release candidate

See the migration guide for an overview of improvements and the docs of the 3 arviz modules: arviz-base, arviz-stats and arviz-plots (all the objects at the top level namespace of these 3 libraries are also exposed as arviz.xyz)

v0.23.0

Important

We expect this to be the last 0.x ArviZ release. See the migration guide to read about what this will mean.

There hasn't been many changes since 0.22, the main ones are fixes in netcdf serialization for InferenceData and fixes to DataTree<->InferenceData conversions.

Full changelog available on GitHub.

v0.22.0

The highlights of these release are interoperability of InferenceData and the new xarray.DataTree object and automatic inferring of dimension names from plates in the numpyro converter.

We have also done more testing of the arviz.preview module. Take a look at the migration guide if you want an early peek to the changes and improvements to come. You can treat arviz.preview as a beta release of ArviZ 1.0 so use carefully and keep an eye on the development. We hope to have a release candidate available for wider user testing in a few months.

Full changelog available on GitHub.

v0.21.0

What's Changed

• handle python -OO by @JohannesBuchner in #2393
• Align autogenerated dimension names when dims and default_dims are provided by @lucianopaz in #2395
• Allow custom groups without warnings by @OriolAbril in #2401
• Robustify preview module by @OriolAbril in #2398
• fix docs right sidebar and bokeh deprecation warning by @OriolAbril in #2405
• add ecdf comparison plot by @OriolAbril in #2178
• Add a parameter to enable/disable smoothing on discrete variables in BLV (Issue #2325) by @Patchouli-Kenntnis in #2344
• Update mcse_sd calculation to not use normality assumption. by @ahartikainen in #2167
• Split up Bayes Factor stat and plotting functions by @mpmbq2 in #2406
• Change Twitter to X, including the icon by @star1327p in #2418
• plot_hdi raise exception when x is string (#2412) by @milesalanmoore in #2413
• DOC: Change Numba links by @star1327p in #2421
• DOC: Corrected typos in ArviZ-Dask integration by @star1327p in #2422
• DOC: Add missing periods to the ArviZ community page by @star1327p in #2426
• DOC: Correct a typo: InfereceData -> InferenceData by @star1327p in #2428
• DOC: Correct a typo in issue_triaging.md by @star1327p in #2431
• Update utils.py by @menacingly-coded in #2430
• DOC: Update two Bokeh reference links by @star1327p in #2432
• DOC: Update Bokeh link in Installation.rst by @star1327p in #2425
• Prepare release by @aloctavodia in #2434

New Contributors

• @JohannesBuchner made their first contribution in #2393
• @Patchouli-Kenntnis made their first contribution in #2344
• @mpmbq2 made their first contribution in #2406
• @star1327p made their first contribution in #2418
• @milesalanmoore made their first contribution in #2413
• @menacingly-coded made their first contribution in #2430

Full Changelog: v0.20.0...v0.21.0

v0.20.0

The highlight this release is the addition of optimized simultaneous ECDF confidence bands. It also includes support for idata["new_group"] = dataset directly and several bug fixes and documentation improvements.

For users of arviz.from_pytree it will now be necessary to install dm-tree manually as it was only used in this one function and has been made an optional dependency.

Full changelog available on GitHub.

v0.19.0

This release highlights are: Work with Bokeh3, uses revised Pareto k threshold, refactor plot_ecdf arguments, expose new features as preview submodule.

Full Changelog: https://github.com/arviz-devs/arviz/blob/main/CHANGELOG.md#v0190-2024-jul-19