Arviz

Maintenance and fixes

- Restructured internals of `from_pymc3` to handle old pymc3 releases and
sliced traces and to provide useful warnings ([1211](https://github.com/arviz-devs/arviz/pull/1211))

Maintenance and fixes

- Fixed bug in `from_pymc3` for sliced `pymc3.MultiTrace` input ([1209](https://github.com/arviz-devs/arviz/pull/1209))

Maintenance and fixes

- Fixed bug in `from_pymc3` when used with PyMC3<3.9 ([1203](https://github.com/arviz-devs/arviz/pull/1203))
- Fixed enforcement of rcParam `plot.max_subplots` in `plot_trace` and
`plot_pair` ([1205](https://github.com/arviz-devs/arviz/pull/1205))
- Removed extra subplot row and column in in `plot_pair` with `marginal=True` ([1205](https://github.com/arviz-devs/arviz/pull/1205))
- Added latest PyMC3 release to CI in addition to using GitHub default branch ([1207](https://github.com/arviz-devs/arviz/pull/1207))

Documentation

- Use `dev` as version indicator in online documentation ([1204](https://github.com/arviz-devs/arviz/pull/1204))

New features

- Stats and plotting functions that provide `var_names` arg can now filter parameters based on partial naming (`filter="like"`) or regular expressions (`filter="regex"`) (see [1154](https://github.com/arviz-devs/arviz/pull/1154)).
- Add `true_values` argument for `plot_pair`. It allows for a scatter plot showing the true values of the variables ([1140](https://github.com/arviz-devs/arviz/pull/1140))
- Allow xarray.Dataarray input for plots.([1120](https://github.com/arviz-devs/arviz/pull/1120))
- Revamped the `hpd` function to make it work with mutidimensional arrays, InferenceData and xarray objects ([1117](https://github.com/arviz-devs/arviz/pull/1117))
- Skip test for optional/extra dependencies when not installed ([1113](https://github.com/arviz-devs/arviz/pull/1113))
- Add option to display rank plots instead of trace ([1134](https://github.com/arviz-devs/arviz/pull/1134))
- Add out-of-sample groups (`predictions` and `predictions_constant_data`) to `from_dict` ([1125](https://github.com/arviz-devs/arviz/pull/1125))
- Add out-of-sample groups (`predictions` and `predictions_constant_data`) and `constant_data` group to pyro and numpyro translation ([1090](https://github.com/arviz-devs/arviz/pull/1090), [1125](https://github.com/arviz-devs/arviz/pull/1125))
- Add `num_chains` and `pred_dims` arguments to from_pyro and from_numpyro ([1090](https://github.com/arviz-devs/arviz/pull/1090), [1125](https://github.com/arviz-devs/arviz/pull/1125))
- Integrate jointplot into pairplot, add point-estimate and overlay of plot kinds ([1079](https://github.com/arviz-devs/arviz/pull/1079))
- New grayscale style. This also add two new cmaps `cet_grey_r` and `cet_grey_r`. These are perceptually uniform gray scale cmaps from colorcet (linear_grey_10_95_c0) ([1164](https://github.com/arviz-devs/arviz/pull/1164))
- Add warmup groups to InferenceData objects, initial support for PyStan ([1126](https://github.com/arviz-devs/arviz/pull/1126)) and PyMC3 ([1171](https://github.com/arviz-devs/arviz/pull/1171))
- `hdi_prob` will not plot hdi if argument `hide` is passed. Previously `credible_interval` would omit HPD if `None` was passed ([1176](https://github.com/arviz-devs/arviz/pull/1176))
- Add `stats.ic_pointwise` rcParam ([1173](https://github.com/arviz-devs/arviz/pull/1173))
- Add `var_name` argument to information criterion calculation: `compare`,
`loo` and `waic` ([1173](https://github.com/arviz-devs/arviz/pull/1173))

Maintenance and fixes

- Fixed `plot_pair` functionality for two variables with bokeh backend ([1179](https://github.com/arviz-devs/arviz/pull/1179))
- Changed `diagonal` argument for `marginals` and fixed `point_estimate_marker_kwargs` in `plot_pair` ([1167](https://github.com/arviz-devs/arviz/pull/1167))
- Fixed behaviour of `credible_interval=None` in `plot_posterior` ([1115](https://github.com/arviz-devs/arviz/pull/1115))
- Fixed hist kind of `plot_dist` with multidimensional input ([1115](https://github.com/arviz-devs/arviz/pull/1115))
- Fixed `TypeError` in `transform` argument of `plot_density` and `plot_forest` when `InferenceData` is a list or tuple ([1121](https://github.com/arviz-devs/arviz/pull/1121))
- Fixed overlaid pairplots issue ([1135](https://github.com/arviz-devs/arviz/pull/1135))
- Update Docker building steps ([1127](https://github.com/arviz-devs/arviz/pull/1127))
- Updated benchmarks and moved to asv_benchmarks/benchmarks ([1142](https://github.com/arviz-devs/arviz/pull/1142))
- Moved `_fast_kde`, `_fast_kde_2d`, `get_bins` and `_sturges_formula` to `numeric_utils` and `get_coords` to `utils` ([1142](https://github.com/arviz-devs/arviz/pull/1142))
- Rank plot: rename `axes` argument to `ax` ([1144](https://github.com/arviz-devs/arviz/pull/1144))
- Added a warning specifying log scale is now the default in compare/loo/waic functions ([1150](https://github.com/arviz-devs/arviz/pull/1150)).
- Fixed bug in `plot_posterior` with rcParam "plot.matplotlib.show" = True ([1151](https://github.com/arviz-devs/arviz/pull/1151))
- Set `fill_last` argument of `plot_kde` to False by default ([1158](https://github.com/arviz-devs/arviz/pull/1158))
- plot_ppc animation: improve docs and error handling ([1162](https://github.com/arviz-devs/arviz/pull/1162))
- Fix import error when wrapped function docstring is empty ([1192](https://github.com/arviz-devs/arviz/pull/1192))
- Fix passing axes to plot_density with several datasets ([1198](https://github.com/arviz-devs/arviz/pull/1198))

Deprecation

- `hpd` function deprecated in favor of `hdi`. `credible_interval` argument replaced by `hdi_prob`throughout with exception of `plot_loo_pit` ([1176](https://github.com/arviz-devs/arviz/pull/1176))
- `plot_hpd` function deprecated in favor of `plot_hdi`. ([1190](https://github.com/arviz-devs/arviz/pull/1190))

Documentation

- Add classifier to `setup.py` including Matplotlib framework ([1133](https://github.com/arviz-devs/arviz/pull/1133))
- Image thumbs generation updated to be Bokeh 2 compatible ([1116](https://github.com/arviz-devs/arviz/pull/1116))
- Add new examples for `plot_pair` ([1110](https://github.com/arviz-devs/arviz/pull/1110))
- Add examples for `psislw` and `r2_score` ([1129](https://github.com/arviz-devs/arviz/pull/1129))
- Add more examples on 2D kde customization ([1158](https://github.com/arviz-devs/arviz/pull/1158))
- Make docs compatible with sphinx3 and configure `intersphinx` for better
references ([1184](https://github.com/arviz-devs/arviz/pull/1184))
- Extend the developer guide and add it to the website ([1184](https://github.com/arviz-devs/arviz/pull/1184))

New features

- Add out-of-sample predictions (`predictions` and `predictions_constant_data` groups) to pymc3, pystan, cmdstan and cmdstanpy translations ([983](https://github.com/arviz-devs/arviz/pull/983), [1032](https://github.com/arviz-devs/arviz/pull/1032) and [1064](https://github.com/arviz-devs/arviz/pull/1064))
- Started adding pointwise log likelihood storage support ([794](https://github.com/arviz-devs/arviz/pull/794), [1044](https://github.com/arviz-devs/arviz/pull/1044) and [1064](https://github.com/arviz-devs/arviz/pull/1064))
- Add out-of-sample predictions (`predictions` and `predictions_constant_data` groups) to pymc3 and pystan translations ([983](https://github.com/arviz-devs/arviz/pull/983) and [1032](https://github.com/arviz-devs/arviz/pull/1032))
- Started adding pointwise log likelihood storage support ([794](https://github.com/arviz-devs/arviz/pull/794), [1044](https://github.com/arviz-devs/arviz/pull/1044))
- Violinplot: rug-plot option ([997](https://github.com/arviz-devs/arviz/pull/997))
- Integrated rcParams `plot.point_estimate` ([994](https://github.com/arviz-devs/arviz/pull/994)), `stats.ic_scale` ([993](https://github.com/arviz-devs/arviz/pull/993)) and `stats.credible_interval` ([1017](https://github.com/arviz-devs/arviz/pull/1017))
- Added `group` argument to `plot_ppc` ([1008](https://github.com/arviz-devs/arviz/pull/1008)), `plot_pair` ([1009](https://github.com/arviz-devs/arviz/pull/1009)) and `plot_joint` ([1012](https://github.com/arviz-devs/arviz/pull/1012))
- Added `transform` argument to `plot_trace`, `plot_forest`, `plot_pair`, `plot_posterior`, `plot_rank`, `plot_parallel`, `plot_violin`,`plot_density`, `plot_joint` ([1036](https://github.com/arviz-devs/arviz/pull/1036))
- Add `skipna` argument to `hpd` and `summary` ([1035](https://github.com/arviz-devs/arviz/pull/1035))
- Added `transform` argument to `plot_trace`, `plot_forest`, `plot_pair`, `plot_posterior`, `plot_rank`, `plot_parallel`, `plot_violin`,`plot_density`, `plot_joint` ([1036](https://github.com/arviz-devs/arviz/pull/1036))
- Add `marker` functionality to `bokeh_plot_elpd` ([1040](https://github.com/arviz-devs/arviz/pull/1040))
- Add `ridgeplot_quantiles` argument to `plot_forest` ([1047](https://github.com/arviz-devs/arviz/pull/1047))
- Added the functionality [interactive legends](https://docs.bokeh.org/en/1.4.0/docs/user_guide/interaction/legends.html) for bokeh plots of `densityplot`, `energyplot`
and `essplot` ([1024](https://github.com/arviz-devs/arviz/pull/1024))
- New defaults for cross validation: `loo` (old: waic) and `log` -scale (old: `deviance` -scale) ([1067](https://github.com/arviz-devs/arviz/pull/1067))
- **Experimental Feature**: Added `arviz.wrappers` module to allow ArviZ to refit the models if necessary ([771](https://github.com/arviz-devs/arviz/pull/771))
- **Experimental Feature**: Added `reloo` function to ArviZ ([771](https://github.com/arviz-devs/arviz/pull/771))
- Added new helper function `matplotlib_kwarg_dealiaser` ([1073](https://github.com/arviz-devs/arviz/pull/1073))
- ArviZ version to InferenceData attributes. ([1086](https://github.com/arviz-devs/arviz/pull/1086))
- Add `log_likelihood` argument to `from_pymc3` ([1082](https://github.com/arviz-devs/arviz/pull/1082))
- Integrated rcParams for `plot.bokeh.layout` and `plot.backend`. ([1089](https://github.com/arviz-devs/arviz/pull/1089))
- Add automatic legends in `plot_trace` with compact=True (matplotlib only) ([1070](https://github.com/arviz-devs/arviz/pull/1070))
- Updated hover information for `plot_pair` with bokeh backend ([1074](https://github.com/arviz-devs/arviz/pull/1074))

Maintenance and fixes

- Fixed bug in density and posterior plot bin computation ([1049](https://github.com/arviz-devs/arviz/pull/1049))
- Fixed bug in density plot ax argument ([1049](https://github.com/arviz-devs/arviz/pull/1049))
- Fixed bug in extracting prior samples for cmdstanpy ([979](https://github.com/arviz-devs/arviz/pull/979))
- Fix erroneous warning in traceplot ([989](https://github.com/arviz-devs/arviz/pull/989))
- Correct bfmi denominator ([991](https://github.com/arviz-devs/arviz/pull/991))
- Removed parallel from jit full ([996](https://github.com/arviz-devs/arviz/pull/996))
- Rename flat_inference_data_to_dict ([1003](https://github.com/arviz-devs/arviz/pull/1003))
- Violinplot: fix histogram ([997](https://github.com/arviz-devs/arviz/pull/997))
- Convert all instances of SyntaxWarning to UserWarning ([1016](https://github.com/arviz-devs/arviz/pull/1016))
- Fix `point_estimate` in `plot_posterior` ([1038](https://github.com/arviz-devs/arviz/pull/1038))
- Fix interpolation `hpd_plot` ([1039](https://github.com/arviz-devs/arviz/pull/1039))
- Fix `io_pymc3.py` to handle models with `potentials` ([1043](https://github.com/arviz-devs/arviz/pull/1043))
- Fix several inconsistencies between schema and `from_pymc3` implementation
in groups `prior`, `prior_predictive` and `observed_data` ([1045](https://github.com/arviz-devs/arviz/pull/1045))
- Stabilize covariance matrix for `plot_kde_2d` ([1075](https://github.com/arviz-devs/arviz/pull/1075))
- Removed extra dim in `prior` data in `from_pyro` ([1071](https://github.com/arviz-devs/arviz/pull/1071))
- Moved CI and docs (build & deploy) to Azure Pipelines and started using codecov ([1080](https://github.com/arviz-devs/arviz/pull/1080))
- Fixed bug in densityplot when variables differ between models ([1096](https://github.com/arviz-devs/arviz/pull/1096))

Deprecation

- `from_pymc3` now requires PyMC3>=3.8

Documentation

- Updated `InferenceData` schema specification (`log_likelihood`,
`predictions` and `predictions_constant_data` groups)
- Clarify the usage of `plot_joint` ([1001](https://github.com/arviz-devs/arviz/pull/1001))
- Added the API link of function to examples ([1013](https://github.com/arviz-devs/arviz/pull/1013))
- Updated PyStan_schema_example to include example of out-of-sample prediction ([1032](https://github.com/arviz-devs/arviz/pull/1032))
- Added example for `concat` method ([1037](https://github.com/arviz-devs/arviz/pull/1037))

New features

- Update for pair_plot divergences can be selected
- Default tools follow global (ArviZ) defaults
- Add interactive legend for a plot, if only two variables are used in pairplot

Maintenance and fixes

- Change `packaging` import from absolute to relative format, explicitly importing `version` function