ENH: Preserve column attrs when doing column selection

[kerrickstaley]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

It would be nice if, when doing column selection on a dataframe, the attrs of the individual series were preserved.

Specifically, this code:

import pandas as pd
df = pd.DataFrame({'a': [1, 2]})
df['a'].attrs['foo'] = 'bar'
print(df[['a']]['a'].attrs)

currently prints

{}

but I would like it to print

{'foo': 'bar'}

Feature Description

N/A

Alternative Solutions

N/A

Additional Context

No response

[jbrockmendel]

I tried to do per-column metadata in a personal project circa 2017 and found it very difficult to do robustly. I'm very skeptical it will be supported here anytime soon. (Though I'm more attrs-skeptical than others in general)

[scott-arne]

Just to comment on this, Series-based attrs disappear on seemingly any operation that copies the DataFrame in version 2.1.4 (not exactly sure when this behavior changed).

df = pd.DataFrame({"MySeries": [1, 2, 3]})
df.MySeries.attrs["metadata"] = "this is important"

# This prints an empty dictionary: {}
print(df.head().MySeries.attrs)

# This still prints the attrs: {'metadata': 'this is important'}
print(df.MySeries.attrs)

I know this wasn't the case previously, I believe even in Pandas 2.x.