A Python Oddity
The Oddity
Let’s say you want to intialize a matrix (a two-dimensional list) in python. It’ll be an NxN matrix of numbers. Just for fun we’ll make all the inital elements 1, instead of 0.
One might think the following would be a sensible one-liner.
# sample.py
def print_mat(matrix):
for row in matrix:
print(row)
matrix = [[1] * 10] * 10
print_mat(matrix)
prints
$ python3 sample.py
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
But, what happens if we modify part of the matrix?
# sample.py
# same print_mat as above
matrix = [[1] * 10] * 10
print("--- BEFORE MOD ---")
print_mat(matrix)
matrix[0][3] = 89
print()
print("--- AFTER MOD ---")
print_mat(matrix)
prints
$ python3 sample.py
--- BEFORE MOD ---
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
--- AFTER MOD ---
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
Oops! That clearly creates a list containing ten of the same list.
(If you are confused, we would normally expect only the first row to have been modified to have the 89, not all of the rows)
Explanation
To better illustrate what’s going on, the above declaration of matrix
matrix = [[1] * 10] * 10
is really the same as
inner_list = [1] * 10
matrix = [inner_list] * 10
Since python variables are really just pointers to objects, we’re making a list of ten of the same pointer to one object, in this case our list of ten ones.
The Correct Way
It’s pretty straight forward to fix this, we just use a list comprehension instead of the * shorthand, which produces this unexpected behavior when applied to a list of lists.
# sample.py
matrix = [[1] * 10 for _ in range(10)]
print("--- BEFORE MOD ---")
print_mat(matrix)
matrix[0][3] = 89
print()
print("--- AFTER MOD ---")
print_mat(matrix)
which produces the expected output
$ python3 sample.py
--- BEFORE MOD ---
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
--- AFTER MOD ---
[1, 1, 1, 89, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
Note: we could also say
matrix = [[1 for _ in range(10)] for _ in range(10)], but that is more to type!