How to Swap Odd and Even Columns in a Pandas DataFrame

Rescuing pandas (16) – Swapping Odd and Even Columns in a DataFrame

Many users avoid pandas for data manipulation, so this tutorial shows how to make pandas more appealing by swapping the positions of odd‑ and even‑indexed columns.

Data Requirement

The task, posed by a community member, is to exchange column A with B, C with D, etc. For an even number of columns this is straightforward; the article also covers the more complex case of an odd number of columns.

import pandas as pd

en = 'abcdefg'
df = pd.DataFrame(([i + j for j in en] for i in en), columns=list(en.upper()), index=list(en.upper()))

Target layout: adjacent odd‑even column pairs should swap, while the final column G remains unchanged if it has no partner.

Requirement Breakdown

First treat the DataFrame as if it had an even number of columns, then handle the odd‑column case separately.

Processing Even‑Column DataFrames

import pandas as pd

en = 'abcdef'
df = pd.DataFrame(([i + j for j in en] for i in en), columns=list(en.upper()), index=list(en.upper()))

Using NumPy, extract odd and even columns, stack them into a 2×n array, and flatten column‑wise:

import numpy as np

odd_c = df.columns[::2]  # odd columns
even_c = df.columns[1::2]  # even columns

# Stack with even columns on top, then flatten column‑wise
new_c = np.array((even_c, odd_c)).flatten('F')
print(df[new_c])

Another approach reshapes the column values into a 2×n matrix and then flattens:

new_c = df.columns.values.reshape((2, len(df.columns) // 2), order='F')

Flipping the rows and flattening also achieves the swap:

df[new_c[::-1].flatten('F')]

Using the power of (‑1)ⁱ, a concise list comprehension swaps adjacent columns:

new_c = [df.columns[i + (-1) ** i for i in range(len(df.columns))]
# ['B', 'A', 'D', 'C', 'F', 'E']
df[new_c]

Handling an Odd Number of Columns

When the column count is odd, separate the last column, apply the even‑column method to the rest, then append the final column back:

import numpy as np

c = df.columns.values
c_ = c[:-1]          # all but last column
c_last = c[-1]       # last column
c_new_even = c_.reshape((2, len(c_) // 2), order='F')[::-1].flatten('F')
new_c = np.hstack([c_new_even, c_last])

A more generic solution works for both even and odd column counts by dynamically determining whether a trailing column exists:

import numpy as np

c = df.columns.values
c_middle = len(c) // 2
c_last = [c[-1]] * (0 + len(c) % 2)
new_c = np.hstack((c[:c_middle * 2].reshape((2, c_middle), order='F')[::-1].flatten('F'), c_last))

When using the (‑1)ⁱ trick on an odd‑column DataFrame, the generated index may exceed the maximum column index; clipping fixes this:

[i + (-1) ** i for i in range(len(df.columns))]

import numpy as np

np.clip([i + (-1) ** i for i in range(len(df.columns))], 0, len(df.columns) - 1)

Summary

This example demonstrates several straightforward ways to reorder odd and even columns in a pandas DataFrame by leveraging column indices, NumPy array reshaping, and the mathematical property of (‑1)ⁱ. The methods are concise, flexible, and illustrate how to think creatively about data‑frame manipulation.

— Written on May 20, 2022