v0.7.0

What's Changed

New features

• Add reloo for exact refitting in PSIS-LOO-CV by @jordandeklerk in #148
• Add method for multiple chains, remove optimization method by @aloctavodia in #151
• Log Weights API change for LOO functions by @jordandeklerk in #156
• Add warnings for finite data and fix computation for discrete data by @aloctavodia in #159
• Add loo_kfold for exact K-fold cross validation by @jordandeklerk in #158
• Compute isotonic regression for categorical and ordinal data by @aloctavodia in #167
• Organize computations for visualizations and make it easy to apply on groupbys by @OriolAbril in #162
• Add functionality to compute quantile dots by @aloctavodia in #171
• Add mode computation by @aloctavodia in #174
• Add support for Jacobian adjustment to elpd_loo values by @jordandeklerk in #175
• Add more informative errors for model comparison in compare() by @jordandeklerk in #184
• Add loo_i function for single observation PSIS-LOO-CV computation by @jordandeklerk in #182
• Add correct output for loo_kfold() by @jordandeklerk in #190
• Implement array interface ecdf by @aloctavodia in #188
• Add function loo_score() for CRPS and SCRPS using PWM identity by @jordandeklerk in #196
• Update default ci_prob value by @aloctavodia in #200
• Add functions to support plot_ppc_censored by @aloctavodia in #203
• Add subsampling support in compare() by @jordandeklerk in #216
• Add order statistic diagnostic for compare() by @jordandeklerk in #237

Maintenance and bug fixes

• Remove extra metrics.py file by @OriolAbril in #152
• Fix function-specific k-hat computation by @jordandeklerk in #165
• Fix loo_moment_match to be more modular by @jordandeklerk in #166
• Bump actions/download-artifact from 4 to 5 by @dependabot[bot] in #173
• Bump actions/checkout from 4 to 5 by @dependabot[bot] in #172
• Bump actions/setup-python from 5 to 6 by @dependabot[bot] in #194
• Fix typing issue in compare() by @jordandeklerk in #198
• Rework PLPD and LPD approximation for loo_subsample() by @jordandeklerk in #192
• Make hypergeom computation more robust by @aloctavodia in #199
• Add callable log likelihood argument to loo_i() by @jordandeklerk in #201
• Robustify pareto_min_ss computation by @aloctavodia in #204
• Survival helper: use view instead of copy by @aloctavodia in #205
• Make pareto_khat consistent with pareto_min_ss by @aloctavodia in #206
• Remove unused function by @aloctavodia in #207
• Add Python 3.13 to tests by @aloctavodia in #209
• Pin numba version by @aloctavodia in #210
• Use a lookup table for hypergeometric computation instead of numba by @aloctavodia in #211
• Fix the sign of PSIS inputs in loo_approximate_posterior() by @jordandeklerk in #213
• Add computational fixes for loo_moment_match() by @jordandeklerk in #212
• Make thin works with arrays by @aloctavodia in #214
• Automatically compute reff for pareto_khat by @aloctavodia in #215
• Patch computations in compare() for subsampling by @jordandeklerk in #217
• Fix stacking weights computation by @aloctavodia in #226
• Add support for Jacobian adjustment to subsampled PSIS-LOO-CV by @jordandeklerk in #227
• Bump actions/download-artifact from 5 to 6 by @dependabot[bot] in #228
• Refactor and expand tests by @jordandeklerk in #220
• Add Jacobian tests for PSIS-LOO-CV subsampling by @jordandeklerk in #231
• Extend CI jobs and tox settings by @OriolAbril in #230
• Remove redundancy in _split_moment_match() by @jordandeklerk in #235

Documentation

• Improve documentation by @jordandeklerk in #197
• Improve the API formatting for arviz_stats.compare by @star1327p in #202
• Remove tutorial section (and duplicated link) by @aloctavodia in #208
• Update the "See Also" parts of API Reference by @star1327p in #221
• Add links to Testing arviz-stats page by @star1327p in #222
• Add link to EABM chapter by @jordandeklerk in #229
• Fix return docstrings for loo functions by @jordandeklerk in #233
• Remove extra references in arviz_stats.rhat by @star1327p in #236
• Fix references and bullet points in arviz_stats.rhat_nested by @star1327p in #239
• Add the reference Säilynoja et al. (2022) from EABM by @star1327p in #241
• Change truncation_factor name to avoid possible confusion by @aloctavodia in #240
• Add tests for numpy logsumexp by @jordandeklerk in #238

New Contributors

• @star1327p made their first contribution in #202

Full Changelog: v0.6.0...v0.7.0