如何用更Pythonic的方式按索引条件切片MultiIndex DataFrame？

Problem Context #

You're working with a MultiIndex pandas DataFrame and need to select two sets of rows:

• All rows where the first-level index is foo
• The single row where first-level index is bar and second-level index is two

Your current approach using pd.concat works, but it's unnecessarily verbose. Here are more concise, Pythonic alternatives:

Method 1: Use pd.IndexSlice (Most Readable) #

Pandas' pd.IndexSlice is purpose-built for MultiIndex operations, making your slicing logic intuitive and clean:

import pandas as pd

# Initialize the IndexSlice helper
idx = pd.IndexSlice

# Combine your filtering conditions in one .loc call
result = df.loc[idx[['bar', 'foo'], ['two', slice(None)]]].sort_index()

Breakdown: #

• idx[['bar', 'foo'], ...]: Targets rows where the first index level is either bar or foo
• ..., ['two', slice(None)]]: Specifies sub-index rules:
  • For bar: only keep the two sub-row
  • For foo: keep all sub-rows (slice(None) acts the same as :)
• sort_index() ensures the output matches your desired row order exactly

Method 2: Directly Pass Target Index Tuples #

If you know the exact index pairs you need, pass them directly to .loc—this is the most straightforward approach for fixed target rows:

# List all (level1, level2) index pairs you want to select
target_rows = [('bar', 'two'), ('foo', 'one'), ('foo', 'two')]
result = df.loc[target_rows]

This method is perfect when you have a small, predefined set of rows to pick.

Method 3: Use query() (Great for Complex Conditions) #

For more intricate filtering logic, query() lets you write conditions in plain, natural language (just name your index levels first):

# Temporarily name your index levels for the query
df.index.names = ['level_1', 'level_2']

# Write your filter condition like a regular boolean statement
result = df.query("(level_1 == 'bar' and level_2 == 'two') or level_1 == 'foo'")

# Optional: revert index names to their original unnamed state
df.index.names = [None, None]

All these methods eliminate the need for pd.concat and align with pandas best practices—they're shorter, easier to read, and maintainable.

内容的提问来源于stack exchange，提问作者crs