Beta release

Highlights

• Added to_dataframe method, __getitem__ magic and copy method to InferenceData
• Add ref_line, bar, vlines and marker_vlines kwargs to plot_rank
• Add observed argument to (un)plot observed data in plot_ppc
• Add support for named dims and coordinates with multivariate observations
• Add support for discrete variables in rank plots and loo_pit
• Make stacking the default method to compute weights in compare (1438)
• [experimental] Modified SamplingWrapper base API

Maintenance and fixes

• Updated CmdStanPy interface
• Have from_pystan store attrs as strings to allow netCDF storage
• Use circular KDE function and fix tick labels in circular plot_trace
• Fix computation of weights in compare
• Add grid argument to manually specify the number of rows and columns
• Switch to compact=True by default in our plots
• Make from_pymc3 compatible with theano-pymc 1.1.0

Deprecations

• plot_khat deprecate annotate argument in favor of threshold.

---

See detailed change log

Beta release

Highlights

• Added support for circular variables to several functions
• Added InferenceData JSON converter and to_dict function
• Added plot_separation for binary data
• Extended InferenceData capabilities for computation and combination of multiple objects.
• Ridgeplot is now hdi aware
• Add support for discrete variables in plot_bpv

Fixes

• Fixed import error with xarray 0.16.1
• Restructured code to be compatible with matplotlib>=3.3

Deprecations

• This will be the last ArviZ version to support Python 3.6. ArviZ will follow NEP 29

---

See detailed change log

Beta release

Highlights

• loo-pit KDE and HDI were improved
• html_repr of InferenceData objects for jupyter notebooks
• Support for PyJAGS
• from_pymc3 automatically retrieves coords and dims from model context
• plot_trace now supports multiple aesthetics to identify chain and variable shape and supports matplotlib aliases
• plot_hdi can now take already computed HDI values

Deprecations

• from_pymc3 without a model context available raises aFutureWarning and will be deprecated in a future version
• In plot_trace, chain_prop and compact_prop as tuples will now raise a FutureWarning
• hdi with 2d data raises a FutureWarning

---

See detailed change log

Beta release

Third patch release on v0.8.0. Includes fixes on from_pymc3

---

See also detailed change log of v0.8.0

Beta release

Second patch release on v0.8.0. Includes fixes on from_pymc3

---

See also detailed change log of v0.8.0

Beta release

Patch release on v0.8.0. Changes are:

Maintenance and fixes

• Fixed bug in from_pymc3 when used with PyMC3<3.9 (#1203)
• Fixed enforcement of rcParam plot.max_subplots in plot_trace and plot_pair (#1205)
• Removed extra subplot row and column in in plot_pair with marginal=True (#1205)
• Added latest PyMC3 release to CI in addition to using GitHub master (#1207)

Documentation

• Use dev as version indicator in online documentation (#1204)

---

See also detailed change log of v0.8.0

Beta release

Highlights

• Extensions to InferenceData:
  • Added support for warmup data
  • Extended support for predictions and log_likelihood groups
  • Added InferenceData.map method
• var_names argument in stats and plotting functions now supports filtering parameters based on partial naming (filter="like") or regular expressions (filter="regex")
• hdi has been extended to work on arrays with more than 2d and on InferenceData objects
• New option in plot_trace to display rank plot

Deprecations

• functions hpd and plot_hpd have been deprecated in favour of hdi and plot_hdi respectively
• argument credible_interval has been deprecated in favour of hdi_prob

---

See detailed change log

Beta release

Highlights

• New defaults for cross validation: loo (old: waic) and log -scale (old: deviance -scale)
• Extensions to InferenceData schema
  • Added support for out of sample posterior predictive with predictions group
  • Added storage support for pointwise log likelihood data from multiple variables with log_likelihood group
• Improved legends: plot_density, plot_energy and plot_ess support interactive legends in Bokeh, automatic legend in plot_trace(..., compact=True) in matplotlib
• Added transform argument to plotting functions
• Better rcParams integration

Experimental features

• Added arviz.wrappers module to allow ArviZ to refit the models if necessary
• Added reloo function to ArviZ

---

See detailed change log

Beta release

Minor release due to error in packaging import statement.

#975

• Update for pair_plot (divergences can be selected
• Default tools follow global (ArviZ) defaults
• Change packaging import from absolute to relative format, explicitly importing version function

#976

• Add interactive legend for a plot, if only two variables are used in pairplot.

Beta release

Highlights

• Initial bokeh support.
• Fully support numpyro (@fehiepsi )
• log_likelihood and observed data from pyro
• ArviZ.jl (@sethaxen )
• improve rcparams
• fix az.concat functionality (@anzelpwj )
• distplot docstring plotting example (@jscarbor )