Replace collections' defaultdict by a normal dict with setdefault

Question

I often used collections.defaultdict to be able to append an element to d[key] without having to initialize it first to [] (benefit: you don't need to do: if key not in d: d[key] = [] ):

import collections, random
d = collections.defaultdict(list)  
for i in range(100):
    j = random.randint(0,20)
    d[j].append(i)  # if d[j] does not exist yet, initialize it to [], so we can use append directly

Now I realize we can simply use a normal dict and setdefault :

import random
d = {}
for i in range(100):
    j = random.randint(0,20)
    d.setdefault(j, []).append(i)

Question: when using a dict whose values are lists, is there a good reason to use a collections.defaultdict instead of the second method (using a simple dict and setdefault ), or are they purely equivalent?

Answer 1

collections.defaultdict is generally more performant, it is optimised exactly for this task and C-implemented. However, you should use dict.setdefault if you want accessing an absent key in your resulting dictionary to result in a KeyError rather than inserting an empty list. This is the most important practical difference.

Answer 2

In addition to the answer by Chris_Rands, I want to further emphasize that a primary reason to use defaultdict is if you want key accesses to always succeed, and to insert the default value if there was none. This can be for any reason, and a completely valid one is the convenience of being able to use [] instead of having to call dict.setdefault before every access.

Also note that key in default_dict will still return False if that key has never been accessed before, so you can still check for existence of keys in a defaultdict if necessary. This allows appending to the lists without checking for their existence, but also checking for the existence of the lists if necessary.

Answer 3

When using defaultdict you have a possibility to do inplace addition:

import collections, random
d = collections.defaultdict(list)  
for i in range(100):
    j = random.randint(0,20)
    d[j] += [i]

There is no equivalent construction like d.setdefault(j, []) += [i] , it gives SyntaxError: cannot assign to function call .