I understand that in general it’s (somewhat) faster to iterate over a list rather than over a set or dict keys. Confirming:

import timeit

list_setup = 'my_list = list(range(100))'
list_stmt = 'for n in my_list: n**2'
print(f'list    : {timeit.timeit(setup=list_setup, stmt=list_stmt, number=100000)}')

set_setup = 'my_set = set(range(100))'
set_stmt = 'for n in my_set: n**2'
print(f'set     : {timeit.timeit(setup=set_setup, stmt=set_stmt, number=100000)}')

dict_setup = 'my_dict = {n: None for n in range(100)}'
dict_stmt = 'for n in my_dict.keys(): n**2'
print(f'dict    : {timeit.timeit(setup=dict_setup, stmt=dict_stmt, number=100000)}')

produces:

list    : 5.110044667
set     : 5.137189292
dict    : 5.179844875000001

But if the values of the dict are objects instead of None, iterating over the dict keys is significantly faster:

dict_obj_setup = '''
class MyClass(object): 
    pass
my_dict_obj = {n: MyClass() for n in range(100)}
'''
dict_obj_stmt = 'for n in my_dict_obj.keys(): n**2'
print(f'dict_obj: {timeit.timeit(setup=dict_obj_setup, stmt=dict_obj_stmt, number=100000)}')

dict_obj: 1.7423309159999985

Why is this happening? And if I have a group of integers that I need to iterate over many, many times, should I create a dict whose keys are my integers and whose values are any random object?!

By the way, the difference goes away if I use n*n instead of n**2 in all of the loops:

list    : 0.32518725
set     : 0.344315166
dict    : 0.3399045409999999
dict_obj: 0.3662140410000001