BUG: reset_index after a group_by raise a ValueError for empty dataframe

[etiennebrateau]

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

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

• I have confirmed this bug exists on the master branch of pandas.

Reproducible Example

import pandas as pd
import datetime as dt

df = pd.DataFrame([(dt.date.today(), "b", 12)], columns=["date", "b", "count"])
df["date"] = pd.to_datetime(df["date"])
df = df[df["count"] == 1]  # uncomment this line to make the dataframe empty and so reset_index raising an exception
df2 = df.set_index('date').groupby(['b']).resample('M').sum().reset_index()

Issue Description

When you run the code above, df won’t be empty and the code will run correctly. But if you uncomment the line that makes the dataframe empty, the reset index will raise a ValueError(f"cannot insert b, already exists").

This is a regression as it was working in pandas 1.2.x

Expected Behavior

It’s expected for the reset_index to apply correctly on an empty dataframe too.

Installed Versions

INSTALLED VERSIONS

commit : 73c6825
python : 3.9.7.final.0
python-bits : 64
OS : Darwin
OS-release : 20.6.0
Version : Darwin Kernel Version 20.6.0: Mon Aug 30 06:12:21 PDT 2021; root:xnu-7195.141.6~3/RELEASE_X86_64
machine : x86_64
processor : i386
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.UTF-8
pandas : 1.3.3
numpy : 1.21.2
pytz : 2021.1
dateutil : 2.8.2
pip : 21.1.2
setuptools : 57.0.0
Cython : None
pytest : 6.2.5
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : 2.9.1 (dt dec pq3 ext lo64)
jinja2 : None
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyxlsb : None
s3fs : None
scipy : 1.7.1
sqlalchemy : 1.4.25
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : None

[rhshadrach]

Thanks for the report! Similar to #44350 since resample is using apply, however we should be able to predict the result shape and correct it in pandas.core.resample.Resampler._wrap_result.

[simonjayhawkins]

This is a regression as it was working in pandas 1.2.x

first bad commit: [3408a61] BUG: Groupby ops on empty objects loses index, columns, dtypes (#39940)

[ahmedibrhm]

I don't think this is a bug but the normal/expected behavior.
The reporter have a DataFrame that's equal to

Date       b  count
7/11/2022 "b"  12

When we apply
def function(df): df.set_index('date').groupby(['b']).resample('M').sum()
The result will be

              count
Date       b
7/11/2022 "b"  12

What we notice here is that column b is dropped because the data is a string not numeric. In this case, we can apply reset_index and it will return the dataframe to its original shape.
What if the DataFrame is as the following

Date       b  count
7/11/2022  13  12

The result of calling the function will be

                b  count
Date       b
7/11/2022  13   13  12

Thus, the b will exist as both Column and Index. Thus, it will cause an error when calling reset_index

The function apply will assume that all of the columns can be numeric when calling the function on the DataFrame when it's empty.
Thus, the expected results should contain b as both Column and Index.
Thus, calling reset_index after it will cause an error as b is both an index and a column.
This behavior is better than dropping the columns that are empty because it will cause the empty data frame to lose all of the columns when applying resample. This was the old behavior that was fixed by #39940 that's why it seems like a regression, but I believe that this isn't a regression/bug and the behavior is consistent.

We can edit the method reset_index when dealing with empty data frames. In this case, reset_index can just ignore the indexes that are already in the columns and only return to the column the indexes that are not there. Then delete the remaining duplicated indexes. -- only when the df is empty. However, this will have inconsistent behavior when passing a df that has duplicate keys as index and column. When it's empty it will return a different df shape than when it has entries. Thus, I don't recommend doing this edit.

[etiennebrateau]

Ok, so you can close the issue then. It’s just that users must be aware of handling empty dataframe before trying to do this kind of call 👍

[rhshadrach]

Thanks for the assessment here @ahmedibrhm! In addition, as part of #46560, the silent dropping of column "b" when it is non-empty will raise in 2.0. So in pandas 2.0, both the empty and non-empty versions of the OP example will raise. This is an improvement because it removes the value-dependent behavior.