statistics.geometric_mean gives incorrect results for mixed int/float inputs

[mdickinson]

BPO	28327
Nosy	@mdickinson, @stevendaprano, @iritkatriel
Files	geometric_mean_int_float.patch geometric_mean_int_float_v2.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/stevendaprano'
closed_at = <Date 2021-08-20.10:02:48.549>
created_at = <Date 2016-10-01.12:18:35.892>
labels = ['type-bug', '3.7']
title = 'statistics.geometric_mean gives incorrect results for mixed int/float inputs'
updated_at = <Date 2021-08-20.10:02:48.548>
user = 'https://github.com/mdickinson'

bugs.python.org fields:

activity = <Date 2021-08-20.10:02:48.548>
actor = 'mark.dickinson'
assignee = 'steven.daprano'
closed = True
closed_date = <Date 2021-08-20.10:02:48.549>
closer = 'mark.dickinson'
components = []
creation = <Date 2016-10-01.12:18:35.892>
creator = 'mark.dickinson'
dependencies = []
files = ['44906', '44908']
hgrepos = []
issue_num = 28327
keywords = ['patch']
message_count = 6.0
messages = ['277808', '277809', '277814', '277815', '399948', '399951']
nosy_count = 3.0
nosy_names = ['mark.dickinson', 'steven.daprano', 'iritkatriel']
pr_nums = []
priority = 'normal'
resolution = 'out of date'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue28327'
versions = ['Python 3.7']

[mdickinson]

The following calculations should all be giving the same result:

>>> import statistics
>>> statistics.geometric_mean([2, 3, 5, 7])
3.80675409583932
>>> statistics.geometric_mean([2, 3, 5, 7.0])
1.6265765616977859
>>> statistics.geometric_mean([2, 3, 5.0, 7.0])
2.4322992790977875
>>> statistics.geometric_mean([2, 3.0, 5.0, 7.0])
3.201085872943679
>>> statistics.geometric_mean([2.0, 3.0, 5.0, 7.0])
3.80675409583932

(Correct result is 3.80675409583932.)

The culprit is this line in statistics._product:

mant, scale = 1, 0  #math.frexp(prod)  # FIXME

... and indeed, we should be starting from prod rather than 1 here. But simply using math.frexp has potential for failure if the accumulated integer product overflows a float.

[mdickinson]

Here's a fix. I was planning to add tests, but as far as I can tell geometric_mean currently has no tests at all. Steve, is that correct? That seems like something that should be fixed before the 3.6 release.

[stevendaprano]

Looks good for me.

Thanks for catching this: I knew it was a bug, but then I ran into the issue that I could no longer build 3.6 before I could fix it, and between that and various issues in the real world I never got back to this.

[mdickinson]

New patch, with a (very slightly) cleaner implementation of _frexp_gen.

[iritkatriel]

I can't reproduce this now:

>>> statistics.geometric_mean([2, 3, 5, 7])
3.80675409583932
>>> statistics.geometric_mean([2, 3, 5, 7.0])
3.80675409583932
>>> statistics.geometric_mean([2, 3, 5.0, 7.0])
3.80675409583932
>>> statistics.geometric_mean([2, 3.0, 5.0, 7.0])
3.80675409583932
>>> statistics.geometric_mean([2.0, 3.0, 5.0, 7.0])
3.80675409583932
>>>

The current geometric_mean was added in PR12638. Is this issue about a previous version?

[mdickinson]

Is this issue about a previous version?

Yep. Sorry for failing to close this earlier.