MAINT, BUG (?): pearsonr statistic return type change about scipy HOT 4 CLOSED

For 1.13, I'll just fix it. I just let a 0d array slip through. I actually try to limit my reflexive use of [()], and in this case I went to far.

I don't think this is in the 1.13 release. gh-20284 will fix this, and the type, dtype, and shape tests are stronger with xp_assert_close, so it won't happen again. In gh-19186 , we specifically added a check for NumPy arrays to ensure that np.float64 is preferred over 0D NumPy arrays:

from scipy.

I probably tested on main.

from scipy.

Here's the result of git bisect process on the above regression test:

ceea3e316ab78e3477004da0c26236105fe3b4d5 is the first bad commit
commit ceea3e316ab78e3477004da0c26236105fe3b4d5
Author: Matt Haberland <[email protected]>
Date:   Wed Mar 20 21:43:07 2024 -0700

    ENH: stats.pearsonr: add support for `axis` argument (#20137)

 scipy/stats/_morestats.py       |  16 ++--
 scipy/stats/_stats_py.py        | 187 ++++++++++++++++++++++++++--------------
 scipy/stats/tests/test_stats.py | 101 ++++++++++++++++++++--
 3 files changed, 219 insertions(+), 85 deletions(-)

from scipy.

Yeah, so @stefanv it looks like this commit is not in the 1.13 branch. Was the problem definitely with 1.13.0rc1, or just main?

from scipy.