• [x] I have checked that this issue has not already been reported.

  • [x] I have confirmed this bug exists on the latest version of pandas.

  • [ ] (optional) I have confirmed this bug exists on the master branch of pandas.

Code Sample, a copy-pastable example

import pandas as pd

s = pd.Series([1, 2, 3])
s1 = s.iloc[1:]
s1 -= 4
s

# output
# 0    1
# 1   -2
# 2   -1
# dtype: int64

s = pd.Series([1, 2, 3])
s1 = s.iloc[1:]
s1 = s1 - 4
s

# output
# 0    1
# 1    2
# 2    3
# dtype: int64

Problem description

Starting with version 1.1.4 when applying either of +=, -=, *= or /= operators on a series s1 that was sliced from original series s, the change propagates to original series s. When using normal assignment operators s1 = s1 - 4 the problem is not present, which leads to inconsistent behavior.

Output of pd.show_versions()

INSTALLED VERSIONS ------------------ commit : 67a3d4241ab84419856b84fc3ebc9abcbe66c6b3 python : 3.7.9.final.0 python-bits : 64 OS : Windows OS-release : 10 Version : 10.0.18362 machine : AMD64 processor : Intel64 Family 6 Model 158 Stepping 13, GenuineIntel byteorder : little LC_ALL : None LANG : None LOCALE : None.None pandas : 1.1.4 numpy : 1.17.2 pytz : 2020.4 dateutil : 2.8.0 pip : 20.3.3 setuptools : 51.0.0.post20201207 Cython : 0.28.5 pytest : 3.8.1 hypothesis : None sphinx : None blosc : None feather : None xlsxwriter : None lxml.etree : None html5lib : None pymysql : None psycopg2 : None jinja2 : 2.11.2 IPython : 7.19.0 pandas_datareader: None bs4 : None bottleneck : None fsspec : None fastparquet : None gcsfs : None matplotlib : 3.3.3 numexpr : None odfpy : None openpyxl : None pandas_gbq : None pyarrow : None pytables : None pyxlsb : None s3fs : None scipy : 1.5.4 sqlalchemy : None tables : None tabulate : None xarray : None xlrd : None xlwt : None numba : None

Comment From: simonjayhawkins

Thanks @nikita-ivanov for the report.

Starting with version 1.1.4 when applying either of +=, -=, *= or /= operators on a series s1 that was sliced from original series s, the change propagates to original series s.

This was changed in [d8c8cbb90544b77e5da1daa7274eba7bef3257b9] REGR: inplace Series op not actually operating inplace (#37508) to fix a regression for DataFrame behaviour.

It looks like is also corrected a long standing bug/inconsistency with the series behaviour.

cc @jbrockmendel

Comment From: nikita-ivanov

@simonjayhawkins thanks for the update. Do I see it correctly that the future behavior would the same as it is now for -= (and similar operations)? More precisely,

s = pd.Series([1, 2, 3])
s1 = s.iloc[1:]
s1 -= 4
s

# output
# 0    1
# 1   -2
# 2   -1
# dtype: int64

s = pd.Series([1, 2, 3])
s1 = s.iloc[1:]
s1 = s1 - 4
s

# output
# 0    1
# 1   -2
# 2   -1
# dtype: int64

Comment From: simonjayhawkins

for the s1 = s1 - 4 case, the expected output remains as 

# 0    1
# 1    2
# 2    3
# dtype: int64

When using normal assignment operators s1 = s1 - 4 the problem is not present, which leads to inconsistent behavior.

in Python, the a = a+1 is not always equivalent to a+=1.

Under the hood, Python calls 'special methods' for the various syntactical constructs. in this case __add__ and __iadd__

if the object, in this case a, implements __iadd__, that will be called. In the case of mutable sequences, a will change in place. However, when a does not implement __iadd__, the expression a+=1 has the same effect as a = a+1. The expression a+1 will be evaluated first, producing a new object, which is then bound to a. In other words, the object bound to a may or may not change, depending on whether a implements __iadd__

In general, for mutable sequences, it is common for __iadd__ to be implemented and that += happens inplace. For immutable sequences, clearly there is no way for that to happen.

A pandas Series is a mutable sequence that implements __iadd__

Comment From: nikita-ivanov

@simonjayhawkins thanks for clarifications. My concern is that prior to 1.1.4 the behavior was the same for both cases. And I did not find change of behavior information in the release notes. However, this change can cause some bugs (the case for me). I shall switch to s1 = s1 - 4, but it might be worthwhile to mention new behavior in the release notes.

Comment From: simonjayhawkins

@nikita-ivanov I agree that a long standing behaviour has been changed without adequate notice.

We will await comment from @jbrockmendel for the way forward here.

I am reluctant to add a regression tag here as I think the new behaviour is more correct and we are not currently planning any more releases in the 1.1.x series.

Comment From: jbrockmendel

I am reluctant to add a regression tag here as I think the new behaviour is more correct

Agreed.

Comment From: MarcoGorelli

I am reluctant to add a regression tag here as I think the new behaviour is more correct

Agreed.

So, should the release notes be clarified, or can this be closed? @jbrockmendel

Comment From: jbrockmendel

If someone wants to flesh out a release note i wont object

Comment From: SophiaTangHart

I'd like to take it and clarify on the release note. I'd like to let you know that this is my first time contributing.

Comment From: MarcoGorelli

Sure, go ahead - here's the contributing guide https://pandas.pydata.org/docs/dev/development/contributing_docstring.html , please do ask (here or on Slack) if anything's not clear

Comment From: SophiaTangHart

@MarcoGorelli, Thank you. I'm reading the contributing guide. I'm also new to Slack. How do I join Pandas Slack group? Do I refer to version 1.1.4 #38519 if I need clarification? Do I need to address certain people? Thanks!

Comment From: MarcoGorelli

Hey - for Slack: https://pandas.pydata.org/docs/dev/development/community.html?highlight=slack#community-slack

I'd suggest just rewording

https://github.com/pandas-dev/pandas/blob/ac648eeaf5c27ab957e8cd284eb7e49a45232f00/doc/source/whatsnew/v1.1.4.rst#L34

, e.g. to

Fixed regression in inplace arithmetic operation (e.g. `+=`) on a Series not updating the parent DataFrame/Series

Comment From: SophiaTangHart

@MarcoGorelli, thank you for directing me to join Slack. And thank you for your suggestion.

I'm not able to open

  • Fixed regression in inplace arithmetic operation on a Series not updating the parent DataFrame (:issue:36373)

to reword it. How can I access the docstring? Do I need to do it through my GitHut account? Thank you.

Comment From: MarcoGorelli

Do I need to do it through my GitHut account?

Yeah you'll need to fork the repo, clone it, checkout a new branch, stage, commit, push, open a pull request - if you read through https://pandas.pydata.org/docs/dev/development/contributing.html it should all be explained, else we're available on Slack to help out

Comment From: SophiaTangHart

take

Comment From: kathleenhang

@SophiaTangHart Hi there, do you mind if I take the issue since it seems to be inactive for some time?

Comment From: SophiaTangHart

Sure. Sorry for taking a while. I've unassigned myself. Thanks.

Comment From: kathleenhang

take