BUG: pandas.pivot_table margins, dropna and observed parameters not producing expected result

[hugotomasf]

Pandas version checks

• 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 main branch of pandas.

Reproducible Example

import pandas as pd

data = {
    'column_A_1': ['A', 'B', 'A', None, 'D', 'B', 'A'],
    'column_A_2': ['G', 'F', 'J', 'J', 'J', 'F', 'G'],
    'column_A_3': ['6602', '7059', '9805', '3080', '8625', '5741', '9685'],
    'column_A_4': ['A', 'B', 'A', None, 'A', None, 'B'],
    'column_A_4': ['X', None, 'Y', None, 'Z', 'X', 'Y'],
    'column_B_1': ['1', '2', '3', '4', '5', '6', '7'],
    'column_C_1': [0, 2, 5, 9, 8, 3, 7],
    'column_C_2': [12, 75, None, 93, 89, 23, 97],
    'column_C_3': [789, 102, 425, 895, None, 795, None],
    'column_C_3': [15886, 49828, None, 9898, 8085, 9707, 8049]
}
df = pd.DataFrame(data)

pd.pivot_table(df, index=['column_A_1', 'column_A_2', 'column_A_3', 'column_A_4'], columns=['column_B_1'], values=['column_C_1', 'column_C_2', 'column_C_3'], aggfunc={'column_C_1': 'max', 'column_C_2': 'min', 'column_C_3': 'count'}, dropna=False, margins=False, observed=True)

Issue Description

I have a huge dataset with similar structure to the example. I want to pivot the table grouping using the columns A as the index, the values of the columns B as the new columns and aggregate the values of the columns C. I want all columns B values to appear as columns, even if the entire column is NaN. This is because I want to coalesce values from multiple columns into one. Therefore, the parameter dropna should be equal to False. But the DataFrame I get has 336 rows with impossible combinations. For example, the first row A, F, 3080, X has the entire row filled with NaNs since this combination does not exist.

This is a problem because with a small dataset I wouldn't mind. But with a fairly large dataset, numpy returns an error because it has reached the maximum list size. While reading the documentation, I noticed the parameter:

I thought this parameter fixed this issue. Playing around with this parameter, it does not affect the result, it only adds a row. Here is a result of combining these two parameters.

dropna=False, margins=False (Too many rows)

dropna=True, margins=False (Missing Column B values)

dropna=False, margins=True (Same as dropna=False, margins=False?)

dropna=True, margins=True (Same as dropna=True, margins=False?)

I also noticed this parameter:

But it is deprecated, and the default value of True seems to be the value that I need. Forcing this parameter to True does not change the result.

Expected Behavior

I expect with the parameter's combination dropna=False, margins=False and observed=True to get all the rows with plausible combinations (like if I was grouping by) and all the columns with column B values and columns C values.

I don't know if this is a bug or if it is the intended way for the pivot table to work and this is an enhancement.

Installed Versions

INSTALLED VERSIONS

commit : 0691c5c
python : 3.10.6
python-bits : 64
OS : Linux
OS-release : 5.10.235-227.919.amzn2.x86_64
Version : #1 SMP Sat Apr 5 16:59:05 UTC 2025
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : None
LOCALE : en_US.UTF-8

pandas : 2.2.3
numpy : 1.26.4
pytz : 2024.1
dateutil : 2.9.0
pip : 24.0
Cython : None
sphinx : 7.2.6
IPython : 8.23.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.12.3
blosc : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : 2024.3.1
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : 3.1.3
lxml.etree : 5.1.0
matplotlib : 3.8.4
numba : 0.59.1
numexpr : 2.9.0
odfpy : None
openpyxl : 3.1.2
pandas_gbq : None
psycopg2 : None
pymysql : None
pyarrow : 15.0.2
pyreadstat : None
pytest : 8.1.1
python-calamine : None
pyxlsb : None
s3fs : None
scipy : 1.13.0
sqlalchemy : 2.0.29
tables : None
tabulate : 0.9.0
xarray : None
xlrd : 2.0.1
xlsxwriter : None
zstandard : 0.22.0
tzdata : 2024.1
qtpy : 2.4.1
pyqt5 : None

[rhshadrach]

Thanks for the report! While the default value of observed is deprecated, that option is not planned for removal. It wasn't to clear to me if this was part of your concern.

I expect with the parameter's combination dropna=False, margins=False and observed=True to get all the rows with plausible combinations (like if I was grouping by) and all the columns with column B values and columns C values.

Running your example with these values, I get a result with columns involving

'column_B_1', 'column_C_1', 'column_C_2', 'column_C_3'

and 336 rows. This appears to have all columns and rows that are expected, can you detail how this differs from your desired result?

One thing I'll mention is that you have duplicate keys in the data dictionary provided. I assume that wasn't intentional.