ENH: In sort methods, accept sequence or dict as key

[gsehrlich]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

Pretty much every time I use the key kwarg of a pandas sort method, I use it to tell pandas the desired order of the unique values. For example:

import pandas as pd

my_index = pd.Index(['one', 'two', 'three', 'three', 'two', 'one', 'three'])
my_index.sort_values(
    key=lambda idx: (idx
        .to_series()
        .rename(index={'one': 1, 'two': 2, 'three': 3})
        .index
    )
)

This is all a bit redundant, and I suspect it's a common use case.

Relatedly, sometimes I want to use a different key function for different columns/levels. In order to do that, I actually need to call the sort method multiple times.

Feature Description

Accept sequences like ['one', 'two', 'three'] specifying the ascending order of the values.

EDIT: Example equivalent to the sort_values call above:

my_index.sort_values(key=['one', 'two', 'three'])

Also accept dictionaries like {'col_1': ['one', 'two', 'three'], 'col_3': my_sort_fn} to sort different columns or levels with different keys in any format.

Alternative Solutions

Make sure all data that I want to lexsort is always an ordered Categorical.

Additional Context

No response

[rhshadrach]

Thanks for the request. You can accomplish this with map, argsort, and take:

my_index = pd.Index(['one', 'two', 'three', 'three', 'two', 'one', 'three'])
print(my_index.take(my_index.map({'one': 1, 'two': 2, 'three': 3}).argsort()))

# Result:
# Index(['one', 'one', 'two', 'two', 'three', 'three', 'three'], dtype='object')

[mroeschke]

Thanks for the issue, but it appears this hasn't gotten traction in a while so closing