DataFrame.groupby.apply returns different results with copy lambda functions.

[mroeschke]

Code Sample, a copy-pastable example if possible

In [32]: df = pd.DataFrame({"g":[1, 2, 2, 2], "a":[1, 2, 3, 4], "b":[5, 6, 7, 8]})
    ...: 

In [33]: df.groupby('g').apply(lambda x: x)
Out[33]: 
   a  b  g
0  1  5  1
1  2  6  2
2  3  7  2
3  4  8  2

In [34]: df.groupby('g').apply(lambda x: x[:])
Out[34]: 
     a  b  g
g           
1 0  1  5  1
2 1  2  6  2
  2  3  7  2
  3  4  8  2

In [35]: df.groupby('g').apply(lambda x: x.copy(deep=False))
Out[35]: 
   a  b  g
0  1  5  1
1  2  6  2
2  3  7  2
3  4  8  2

In [36]: df.groupby('g').apply(lambda x: x.copy(deep=True))
Out[36]: 
     a  b  g
g           
1 0  1  5  1
2 1  2  6  2
  2  3  7  2
  3  4  8  2

Problem description

xref #9867

Issue #9946 (In[36]) no longer raises a ValueError on master 0.19.1, but I am curious why these 4 cases return 2 different results.

Expected Output

Not sure since I don't have a personal use-case for this operation, but I'd expect these operations to return the same result unless I am missing something with using a deepcopy.

Output of pd.show_versions()

```

commit: 3ccb501
python: 2.7.12.final.0
python-bits: 64
OS: Linux
OS-release: 4.4.0-45-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: None.None

pandas: 0.19.0+228.g3ccb501
nose: 1.3.7
pip: 8.1.2
setuptools: 27.2.0
Cython: 0.24.1
numpy: 1.11.2
scipy: None
statsmodels: None
xarray: None
IPython: 5.1.0
sphinx: 1.4.8
patsy: None
dateutil: 2.5.3
pytz: 2016.7
blosc: None
bottleneck: None
tables: None
numexpr: None
matplotlib: 1.5.3
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.8
s3fs: None
pandas_datareader: None

</details>

[jreback]

so these are examples where things seem to be mutated, the only detection is that the reference changed. This is an implementation detail; we have had a couple of reports very similiar to this.

#14810 (and some others, but this is hard to search for actually).

If you want to try to figure out how to make this better, by all means.

[jorisvandenbossche]

I think we should try to deprecate this behaviour of trying to detect whether the original object is mutated (and then do a transform like operation) before 1.0

[TomAugspurger]

This is being reverted in #35306 and will be re-fixed in #34998 in pandas 1.2.

[MoritzLaurer]

I'm still encountering this issue of inconsistent return formats with pandas 1.4.4 and 1.5
This issue is actually quite frustrating. it makes code with groupby defacto not reproducible and unstable because groupby seems to randomly return different output formats which means that my code sometimes runs and sometimes doesnt :(
see also #13056

e.g with this code the results are inconsistent and I don't understand why df_cl.groupby(by=group).apply(lambda x: x.label.value_counts(sort=True))