如何对列名为四元组的DataFrame按元组元素条件执行groupby分组？

Hey there! Let's start by cleaning up that DataFrame creation first—four nested loops get the job done, but there's a much cleaner (and efficient) way using itertools.product. Then we'll tackle that column grouping problem you're facing with those tuple/namedtuple column names.

Step 1: Optimize DataFrame Creation #

Instead of nested loops, use itertools.product to generate all combinations of your 4 keys in one go. This is way more concise and avoids unnecessary loop overhead:

import numpy as np
import pandas as pd
from collections import namedtuple
from itertools import product

# Define your named tuple structure
TupleKey = namedtuple("TupleKey", ["key1", "key2", "key3", "key4"])

# Replace these with your actual key value lists
key1_options = ["A", "B", "C"]
key2_options = ["X", "Y"]
key3_options = ["P", "Q"]
key4_options = ["M", "N"]

# Generate all possible key combinations
all_column_keys = [TupleKey(*combo) for combo in product(key1_options, key2_options, key3_options, key4_options)]

# Create DataFrame with random data (adjust rows/columns as needed)
random_data = np.random.rand(15, len(all_column_keys))  # 15 rows, 1 column per key combo
df = pd.DataFrame(random_data, columns=all_column_keys)

Step 2: Group Columns by Partial Tuple Elements #

Since your column names are namedtuple instances, you can access their attributes directly (like col.key1) to create grouping keys. Here are common scenarios:

Scenario 1: Group by specific named tuple fields (e.g., key1 and key3) #

If you want to group columns based on fixed tuple elements (regardless of order in your grouping logic), define a lambda function to extract those fields:

# Group columns by key1 and key3 values
grouping_key = lambda col: (col.key1, col.key3)

# Group along columns axis (axis=1) and aggregate (e.g., mean, sum, max)
grouped_df = df.groupby(grouping_key, axis=1).mean()

# The resulting columns will be tuples of (key1, key3) pairs
print(grouped_df.columns)

Scenario 2: Group by unordered tuple elements #

If you need to group columns based on a set of values (ignoring which tuple position they're in), create a sorted tuple of those values to ensure consistent grouping:

# Example: Group columns that contain "X" and "M" in any tuple position
target_values = {"X", "M"}
grouping_key = lambda col: tuple(sorted(target_values.intersection(col)))

# Group and aggregate (using sum here as an alternative)
grouped_df = df.groupby(grouping_key, axis=1).sum()

Scenario 3: Using regular tuples instead of namedtuples #

If you had regular tuples as column names instead of namedtuples, just use index positions instead of attribute names:

# Group by first and third elements of regular tuples
grouping_key = lambda col: (col[0], col[2])
grouped_df = df.groupby(grouping_key, axis=1).median()

Why This Works #

By passing a custom grouping function to df.groupby(axis=1), we're telling pandas to group columns based on the output of that function for each column name. Using namedtuples makes the code more readable than indexing regular tuples, but both approaches work seamlessly.

内容的提问来源于stack exchange，提问作者lanadaquenada