This data as json
| sha | message | author_date | committer_date | raw_author | raw_committer | repo | author | committer |
|---|---|---|---|---|---|---|---|---|
| d61efb687a9b2989ad22c70334b8cc2e32251d16 | Fix behaviour of min_count in reducing functions (#4911) * Add more tests for reducing functions with min_count * Make sure Dask-backed arrays are not computed. * Check some specific examples give the correct output. * Run membership tests on xarray.core.dtypes.NAT_TYPES * Fix behaviour of min_count in reducing functions. * Fix mask checks in xarray.core.nanops._maybe_null_out to run lazily for Dask-backed arrays. * Change xarray.core.dtypes.NAT_TYPES to a set (it is only used for membership checks). * Add dtypes to NAT_TYPES rather than instances. Previously np.float64 was returning true from `dtype in NAT_TYPES` which resulted in min_count being ignored when reducing over all axes. * Add whatsnew entry. * Improvements from review. * use duck_array_ops.where instead of np.where * add docstring and whatsnew messages about sum/prod on integer arrays with skipna=True and min_count != None now returning a float array. Co-authored-by: Deepak Cherian <dcherian@users.noreply.github.com> | 2021-02-19T08:12:01Z | 2021-02-19T08:12:01Z | c67e67f6d6f19d5cc9bcf98a82af5f6a678d899a | cd792325681cbad9f663f2879d8b69f1edbb678f | 13221727 | 367900 | 19864447 |