v0.11.4 (2021 Oct 3)
Maintenance and fixes
* Fix standard deviation code in density utils by replacing it with `np.std`.
v0.11.3 (2021 Oct 1)
New features
* Added `labeller` argument to enable label customization in plots and summary
* Added `arviz.labels` module with classes and utilities
* Added probability estimate within ROPE in `plot_posterior`
* Added `rope_color` and `ref_val_color` arguments to `plot_posterior`
* Improved retrieving or pointwise log likelihood in `from_cmdstanpy`, `from_cmdstan` and `from_pystan`
* Added interactive legend to bokeh `forestplot`
* Added interactive legend to bokeh `ppcplot`
* Add more helpful error message for HDF5 problems reading `InferenceData` from NetCDF
* Added `data.log_likelihood`, `stats.ic_compare_method` and `plot.density_kind` to `rcParams`
* Improve error messages in `stats.compare()`, and `var_name` parameter.
* Added ability to plot HDI contours to `plot_kde` with the new `hdi_probs` parameter.
* Add dtype parsing and setting in all Stan converters
* Add option to specify colors for each element in ppc_plot
Maintenance and fixes
* Fix conversion for numpyro models with ImproperUniform latent sites
* Fixed conversion of Pyro output fit using GPUs
* Enforced using coordinate values as default labels
* Integrate `index_origin` with all the library
* Fix pareto k threshold typo in reloo function
* Preserve shape from Stan code in `from_cmdstanpy`
* Updated `from_pystan` converters to follow schema convention
* Used generator instead of list wherever possible
* Correctly use chain index when constructing PyMC3 `DefaultTrace` in `from_pymc3`
* Fix bugs in CmdStanPyConverter
* Fix `c` argument in `plot_khat`
* Fix `ax` argument in `plot_elpd`
* Remove warning in `stats.py` compare function
* Fix `ess/rhat` plots in `plot_forest`
* Fix `from_numpyro` crash when importing model with `thinning=x` for `x > 1`
* Upload updated mypy.ini in ci if mypy copilot fails
* Added type checking to raise an error whenever `InferenceData` object is passed using `io_pymc3`'s `trace` argument
* Fix `xlabels` in `plot_elpd`
* Renamed `sample` dim to `__sample__` when stacking `chain` and `draw` to avoid dimension collision
* Removed the `circular` argument in `plot_dist` in favor of `is_circular`
* Fix `legend` argument in `plot_separation`
* Removed testing dependency on http download for radon dataset
* Fixed plot_kde to take labels with kwargs.
* Fixed xarray related tests.
* Fix Bokeh deprecation warnings
* Fix credible inteval percentage in legend in `plot_loo_pit`
* Arguments `filter_vars` and `filter_groups` now raise `ValueError` if illegal arguments are passed
* Remove constrained_layout from arviz rcparams
* Fix plot_elpd for a single outlier
Deprecation
* Deprecated `index_origin` and `order` arguments in `az.summary`
Documentation
* Language improvements of the first third of the "Label guide"
* Added "Label guide" page and API section for `arviz.labels` module
* Add "Installation guide" page to the documentation
* Improve documentation on experimental `SamplingWrapper` classes
* Added example to `plot_hdi` using Inference Data
* Removed `geweke` diagnostic from `numba` user guide
* Restructured the documentation sections to improve community and about us information
v0.11.2 (2021 Feb 21)
New features
* Added `to_zarr` and `from_zarr` methods to InferenceData
* Added confidence interval band to auto-correlation plot
Maintenance and fixes
* Updated CmdStanPy converter form compatibility with versions >=0.9.68
* Updated `from_cmdstanpy`, `from_cmdstan`, `from_numpyro` and `from_pymc3` converters to follow schema convention
* Fix calculation of mode as point estimate
* Remove variable name from legend in posterior predictive plot
* Added significant digits formatter to round rope values
* Updated `from_cmdstan`. csv reader, dtype problem fixed and dtype kwarg added for manual dtype casting
Deprecation
* Removed Geweke diagnostic
* Removed credible_interval and include_circ arguments
Documentation
* Added an example for converting dataframe to InferenceData
* Added example for `coords` argument in `plot_posterior` docstring
v0.11.1 (2021 Feb 2)
Maintenance and fixes
* Fixed ovelapping titles and repeating warnings on circular traceplot
* Removed repetitive variable names from forest plots of multivariate variables
* Fixed regression in `plot_pair` labels that prevented coord names to be shown when necessary
Documentation
* Use tabs in ArviZ example gallery
v0.11.0 (2021 Dec 17)
New features
* Added `to_dataframe` method to InferenceData
* Added `__getitem__` magic to InferenceData
* Added group argument to summary
* 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
`loo_pit`
* Add `skipna` argument to `plot_posterior`
* Make stacking the default method to compute weights in `compare`
* Add `copy()` method to `InferenceData` class.
Maintenance and fixes
* prevent wrapping group names in InferenceData repr_html
* Updated CmdStanPy interface
* Remove left out warning about default IC scale in `compare`
* Fixed a typo found in an error message raised in `distplot.py`
* Fix typo in `loo_pit` extraction of log likelihood
* Have `from_pystan` store attrs as strings to allow netCDF storage
* Remove ticks and spines in `plot_violin`
* Use circular KDE function and fix tick labels in circular `plot_trace`
* Fix `pair_plot` for mixed discrete and continuous variables
* Fix in-sample deviance in `plot_compare`
* Fix computation of weights in compare
* Avoid repeated warning in summary
* Fix hdi failure with boolean array
* Automatically get the current axes instance for `plt_kde`, `plot_dist` and `plot_hdi`
* Add grid argument to manually specify the number of rows and columns
* Switch to `compact=True` by default in our plots
* `plot_elpd`, avoid modifying the input dict
* Do not plot divergences in `plot_trace` when `kind=rank_vlines` or `kind=rank_bars`
* Allow ignoring `observed` argument of `pymc3.DensityDist` in `from_pymc3`
* Make `from_pymc3` compatible with theano-pymc 1.1.0
* Improve typing hints
Deprecation
* `plot_khat` deprecate `annotate` argument in favor of `threshold`. The new argument accepts floats
Documentation
* Reorganize documentation and change sphinx theme
* Switch to [MyST](https://myst-parser.readthedocs.io/en/latest/) and [MyST-NB](https://myst-nb.readthedocs.io/en/latest/index.html)
for markdown/notebook parsing in docs
* Incorporated `input_core_dims` in `hdi` and `plot_hdi` docstrings
* Add documentation pages about experimental `SamplingWrapper`s usage
* Show example titles in gallery page
* Add `sample_stats` naming convention to the InferenceData schema
* Extend api documentation about `InferenceData` methods
Experimental
* Modified `SamplingWrapper` base API
v0.10.0 (2020 Sep 24)
New features
* Added InferenceData dataset containing circular variables
* Added `is_circular` argument to `plot_dist` and `plot_kde` allowing for a circular histogram (Matplotlib, Bokeh) or 1D KDE plot (Matplotlib).
* Added `to_dict` method for InferenceData object
* Added `circ_var_names` argument to `plot_trace` allowing for circular traceplot (Matplotlib)
* Ridgeplot is hdi aware. By default displays truncated densities at the specified `hdi_prop` level
* Added `plot_separation`
* Extended methods from `xr.Dataset` to `InferenceData`
* Add `extend` and `add_groups` to `InferenceData`
* Added `__iter__` method (`.items`) for InferenceData
* Add support for discrete variables in `plot_bpv`
Maintenance and fixes
* Automatic conversion of list/tuple to numpy array in distplot
* `plot_posterior` fix overlap of hdi and rope
* `plot_dist` bins argument error fixed
* Improve handling of circular variables in `az.summary`
* Removed change of default warning in `ELPDData` string representation
* Update `radon` example dataset to current InferenceData schema specification
* Update `from_cmdstan` functionality and add warmup groups
* Restructure plotting code to be compatible with mpl>=3.3
* Replaced `_fast_kde()` with `kde()` which now also supports circular variables via the argument `circular`
* Increased `from_pystan` attrs information content
* Allow `plot_trace` to return and accept axes
* Update diagnostics to be on par with posterior package
* Use method="average" in `scipy.stats.rankdata`
* Add more `plot_parallel` examples
* Bump minimum xarray version to 0.16.1
* Fix multi rope for `plot_forest`
* Bump minimum xarray version to 0.16.1
* `from_dict` will now store warmup groups even with the main group missing
* increase robustness for repr_html handling
600a3642f2