Changelog History

xskillscore v0.0.29 (2026-02-18)

Internal Changes

• Lifted numpy<2.4 dependency constraint and bumped minimum to numpy>=1.25. (GH444) Aaron Spring

xskillscore v0.0.28 (2026-01-19)

Bug Fixes

• Fixed a bug introduced in v0.0.27 where numpy v1.x support was broken for some algorithms. (GH431) Trevor James Smith.

• Removed all numpy.atleast_1d() calls that were causing numerical differences in p-value calculations with numpy v2.x and fixed several doctest expected outputs. (GH440) Aaron Spring.

Internal Changes

• Pinned numpy below v2.4 due to breaking API changes. (GH441) Trevor James Smith

• Adjusted tests to adapt to new xarray attributes preservation behaviour. (GH441) Trevor James Smith

• Updated pre-commit configuration. (GH434)

xskillscore v0.0.27 (2025-07-14)

Bug Fixes

• Updated and corrected the build configurations in the GitHub workflows so that the correct Python is used when running automated build tests. (GH426) Trevor James Smith

Internal Changes

• Adapted code base for modern numpy and xarray. (GH426) Trevor James Smith

• Removed xskillscore.core.utils.suppress_warnings() in lieu of warnings.filterwarnings(). (GH426) Trevor James Smith

• The minimum supported versions for several dependencies have been updated. (GH426) Trevor James Smith

• xskillscore now uses setuptools-scm to automatically determine the version number. (GH427) Trevor James Smith

• Updated GitHub Actions and pre-commit hook versions to more modern versions. (GH428) Trevor James Smith

• Release documentation steps have been updated to to refer to the new setuptools-scm-based approach. (GH428) Trevor James Smith

• The test suite now refers to the xarray.testing submodule for basic testing helpers. (GH422) weipeng1999

xskillscore v0.0.26 (2024-03-10)

Internal Changes

• Fix build Ray Bell.

xskillscore v0.0.25 (2024-03-10)

Bug Fixes

• rank_histogram() random_for_tied=True handles tied ranks correctly by default. random_for_tied=False ignores this and retains previous behaviour. (GH335, GH364) Aaron Spring.

• Allow singleton dimension in resample_iterations_idx() as this is allowed in resample_iterations() also. (GH375, GH376) Aaron Spring.

Internal Changes

• Reduce dependencies (GH359, GH363) Aaron Spring.

• Implement typing for all xs.{calls}. (GH317, GH366) Aaron Spring.

• Refactor masking NaN in rps(). (GH397) Aaron Spring.

• Use to pyproject.toml and use ruff for linting Ray Bell.

xskillscore v0.0.24 (2021-10-08)

Documentation

• Replaced Boston house data in tabular.ipynb (GH352) Ray Bell.

xskillscore v0.0.23 (2021-08-09)

Features

• multipletests() controlling the false discovery rate for multiple hypothesis tests. (GH365, GH370) Aaron Spring.

Bug Fixes

• crps_ensemble() broadcasts (GH345, GH346) Aaron Spring.

Internal Changes

• resample_iterations_idx() do not break when dim is not coordinate. (GH303, GH339) Aaron Spring.

• Allow float or integer forecasts in brier_score() (GH285, GH341) Aaron Spring.

xskillscore v0.0.22 (2021-06-29)

Internal Changes

• Created np_probabilistic.py (GH333) Ray Bell.

• Require xhistogram>=0.3.0 (GH337).

xskillscore v0.0.21 (2021-06-13)

• Allow float or integer forecasts in brier_score() (GH285, GH342) Aaron Spring

Internal Changes

• Added mypy to linting (GH320) Zachary Blackwood.

Documentation

• Added more info in quick-start.ipynb (GH316) Ray Bell.

• Created tabular-data.ipynb (GH330) Ray Bell.

Breaking changes

• Renamed mae_test to halfwidth_ci_test to make this comparative metric generic. Now, it accepts any of the distance metrics functions except for mape. The new function has an additional argument called metric which is a string and name of the target distance metric. (GH319) Taher Chegini.

xskillscore v0.0.20 (2021-05-08)

Features

• Specify category distribution type with input_distributions in rps() if category_edges==None that forecasts and observations are probability distributions p or cumulative distributionss c. See rps() docstrings and doctests for examples. (GH300) Aaron Spring

• Added slope of linear fit linslope(). (GH307) Ray Bell

Internal Changes

• Use pytest-xdist and matplotlib-base in environments to speed up CI. (GH283) Aaron Spring

• rps() does not break from masking NaNs anymore. rps() expilicty checks for bin_dim if category_edges==None. (GH287) Aaron Spring

• Add doctest on the docstring examples. (GH302) Ray Bell

• Removed a call to compute weights in testing. (GH306) Ray Bell

• Use built in xarray clip method. (GH309) Ray Bell

xskillscore v0.0.19 (2021-03-12)

Features

• Added mean error me(). (GH202, GH200) Andrew Huang

• brier_score() and rps() now contain keyword fair to account for ensemble-size adjustments, but defaults to False. brier_score() also accepts binary or boolean forecasts when a member_dim dimension is present. (GH162, GH211) Aaron Spring

• Added MAE significance test mae_test() from Jolliffe and Ebert https://www.cawcr.gov.au/projects/verification/CIdiff/FAQ-CIdiff.html (GH192, GH209) Aaron Spring

• resample_iterations() and faster resample_iterations_idx() for resampling with and without replacement. (GH215, GH225) Aaron Spring

• Added receiver operating characteristic (ROC) roc(). (GH114, GH256, GH236, GH259) Aaron Spring

• Added many options for category_edges in rps(), which allows multi-dimensional edges. rps() now requires dimension member_dim in forecasts. (GH275, GH277) Aaron Spring

Breaking changes

• Aligned output of sign_test() with mae_test(). Now tests from comparative.py return more than one object including a boolean indicating signficance based on alpha. (GH209) Aaron Spring

• Drop support for python 3.6. (GH237, GH276) Ray Bell

Bug Fixes

• sign_test() now works for xr.Dataset inputs. (GH198, GH199) Aaron Spring

• threshold_brier_score() does not average over thresholds when dim==None. Now also carries threshold as coordinate. (GH255, GH211) Aaron Spring

• Passing weights no longer triggers eager computation. (GH218, GH224). Andrew Huang

• rps() not restricted to [0, 1]. (GH266, GH277) Aaron Spring

Internal Changes

• Added Python 3.7 and Python 3.8 to the CI. Use the latest version of Python 3 for development. (GH21, GH189) Aaron Spring

• Lint with the latest black. (GH179, GH191) Ray Bell

• Update mape algorithm from scikit-learn v0.24.0 and test against it. (GH160, GH230) Ray Bell

• Pin numba to >=0.52 to fix CI (GH233, GH234) Ray Bell

• Refactor asv benchmarks. (GH231) Aaron Spring

• Added tests for nans in correlation metrics (GH246, GH247) Ray Bell

• Added tests for weighted metrics against scikit-learn (GH257) Ray Bell

• Pin xhistogram to >=0.1.2 and adjust code/documentation so that, as in np.histogram, right-most bin is right-edge inclusive where bins are specified (GH269) Dougie Squire

• Reduce warnings. (GH41, GH268) Aaron Spring

• Use raise_if_dask_computes from xarray. (GH272, GH273) Ray Bell

• threshold_brier_score() now carries threshold values as coordinates. (GH279) Aaron Spring

xskillscore v0.0.18 (2020-09-23)

Features

• Added the sign test described in DelSole and Tippett 2016: sign_test(). (GH133, GH176) Aaron Spring and Dougie Squire

Internal Changes

• Removed an unused variable in _rmse, resulting in 2x speedup (GH182). Andrew Huang

• Require xarray=0.16.1 (GH183, GH184) Aaron Spring

Bug Fixes

• Fix incompatibility with xarray=0.16.1 in apply_ufunc (GH183, GH184) Aaron Spring

Documentation

• Added CONTRIBUTING.md to trigger built-in Github contribution guide reference (GH181) mcsitter.

xskillscore v0.0.17 (2020-09-06)

Features

• Added contingency table Contingency() and associated metrics (GH119, GH153). Dougie Squire

• Added rank_histogram() and discrimination() to probabilistic metrics (GH136). Dougie Squire

• Added reliability() to probabilistic metrics (GH164). Dougie Squire

• Added dim and weights kwargs for probabilistic metrics. (GH121) Aaron Spring

• Added keep_attrs kwarg for all metrics. (GH122) Andrew Huang

• Added ranked probability score rps(). (GH163) Aaron Spring

• Deterministic metrics now automatically broadcast any non-core dimensions. E.g., a single time series can be compared to a gridded product spanning that same time span. (GH165, GH71, GH156, GH166) Aaron Spring

Breaking Changes

• Renamed dim to member_dim in probabilistic metrics. (GH121) Aaron Spring

• Argument dim becomes keyword dim=None in all metrics. (GH137, GH143) Aaron Spring

• dim=None reduces all dimensions as in xr.mean(dim=None). (GH137, GH143) Aaron Spring

Bug Fixes

• Fixes weights=None type issue with latest version of dask. (GH168, GH171) Andrew Huang

Documentation

• Added sphinx documentation with full API and a quick start notebook. (GH127) Riley X. Brady and Ray Bell.

Internal Changes

• Added utils module to house utilities shared across multiple modules (GH119). Dougie Squire

• Added conftest.py to gather all pytest.fixtures. (GH126, GH159). Aaron Spring and Ray Bell

• Removed test_np_deterministic covered by test_metric_results_accurate. (GH159) Aaron Spring

xskillscore v0.0.16 (2020-07-18)

Internal Changes

• Add community support documents: HOWTOCONTRIBUTE.rst, issue template and pull request template. Aaron Spring and Ray Bell

• Replace pandas with cftime in examples and tests. Aaron Spring and Ray Bell

• Add coveralls for tests coverage. Aaron Spring and Ray Bell

• Add black, flake8, isort, doc8 and pre-commit for formatting similar to climpred. Aaron Spring and Ray Bell

Bug Fixes

• Avoid mutating inputted arrays when skipna=True. (GH111) Riley X. Brady.

• Avoid read-only error that appeared due to not copying input arrays when dealing with NaNs. (GH111) Riley X. Brady.

xskillscore v0.0.15 (2020-03-24)

Features

• Update the XSkillScoreAccessor with all metrics. Ray Bell

xskillscore v0.0.14 (2020-03-20)

Features

• Add r2 as an implementation of sklearn.metrics.r2_score. Ray Bell

xskillscore v0.0.13 (2020-03-17)

Bug Fixes

• Fixes https://github.com/xarray-contrib/xskillscore/issues/79 assignment destination is read-only error when skipna=True and weights are passed. Andrew Huang

xskillscore v0.0.12 (2020-01-09)

Internal Changes

• ~30-50% speedup for deterministic metrics when weights=None. Aaron Spring

xskillscore v0.0.11 (2020-01-06)

Features

• Add effective_sample_size, pearson_r_eff_p_value, and spearman_r_eff_p_value for computing statistical significance for temporally correlated data with autocorrelation. Riley X. Brady

xskillscore v0.0.10 (2019-12-21)

Deprecations

• mad no longer works and is replaced by median_absolute_error. Riley X. Brady

Bug Fixes

• skipna for pearson_r and spearman_r and their p-values now reports accurate results when there are pairwise nans (i.e., nans that occur in different indices in a and b) Riley X. Brady

Testing

• Test that results from grid cells in a gridded product match the same value if their time series were input directly into functions. Riley X. Brady

• Test that metric results from xskillscore are the same value as an external package (e.g. numpy, scipy, sklearn). Riley X. Brady

• Test that skipna=True works properly with pairwise nans. Riley X. Brady