How can I dynamically sum paired columns in a large Pandas DataFrame?

Below is one efficient way to achieve this without hardcoding every combination. The idea is to note that each new column f_i should sum all existing columns where i appears (either as the first or second element in the column name). Assuming all column names are in the format "i_j" with i < j, you can do the following:

PYTHONCopied
import pandas as pd

# Assume df is your DataFrame. For example:
data = {
    "1_2": [1, 4, 8, 4, 8],
    "1_3": [5, 3, 8, 3, 0],
    "1_4": [2, 4, 8, 4, 7],
    "2_3": [8, 5, 9, 4, 4],
    "2_4": [2, 8, 3, 8, 2],
    "3_4": [8, 5, 3, 3, 2],
}
df = pd.DataFrame(data)

# Extract the unique numbers that appear in the column names
numbers = set()
for col in df.columns:
    a, b = col.split('_')
    numbers.update([int(a), int(b)])
numbers = sorted(numbers)

# For each number, find all columns where that number appears
for i in numbers:
    relevant_cols = [
        col for col in df.columns 
        if str(i) in col.split('_')
    ]
    df[f"f_{i}"] = df[relevant_cols].sum(axis=1)

print(df)

Explanation

1. Extract Unique Numbers:
   We loop through each column name, split by the underscore, and collect both parts into a set so we know all the individual numbers involved (e.g., 1, 2, 3, 4).

2. Generate f_i Columns:
   For each unique number i, we build a list of columns where i appears (using a list comprehension). Then, we sum those columns row-wise using df[relevant_cols].sum(axis=1) and assign the result to a new column named f_i.

This approach scales gracefully because it automatically deals with the combinations regardless of how many numbers (or columns) you have.