DEV Community

SpeedUp Python List and Dictionary

Prashant Sharma on May 03, 2020

Today, we will be discussing the optimization technique in Python. In this article, you will get to know to speed up your code by avoiding the re-e...

Read full post

Dimitris R • May 3 '20

Wow thanks! I had no idea this was even possible.

For the list append, is it because of the use of len()? Just checked the code because I wanted to know why this is happening. Couldn't figure out the dictionary though.

Dave Cridland • May 4 '20

Both are actually the same effect. When you call data.get(...), you're calling the method, of course, but also looking up the method in the data object's internal dictionary, called __dict__. What the above article is showing is that there's some savings to be made if you cache the lookup.

He's not optimizing a list at all, but he is optimizing a dict - in both cases.

Because data[num] = ... is actually data.set(num, ...), there's another candidate for optimization there, as well - but by now you can probably guess what it is.

panda • May 4 '20

It's was fixed in python3.8

milan997 • May 4 '20

Do you have some source on this? How it was fixed?

Rafael Acioly • May 4 '20

so v3.8 is faster? (like the article says)

Idan Atias • May 4 '20 • Edited

Thanks for sharing.
I wonder though if this reduces code readability.
If we are not talking about mission critical paths, then it may not worth the 10-15 % reduction.

Dimitri Merejkowsky • May 6 '20

I wonder though if this reduces code readability.

Yes it does - idiomatic Python code is typically not written this way.

If we are not talking about mission critical paths, then it may not worth the 10-15 % reduction.

This. Before applying this technique, make sure to measure the performance of your entire application!

Finally, a better advice for beginners (in my opinion), is to advise them to use comprehensions when they can:

new_list = [append(num) for num in list]

is more idiomatic and even faster than the technique presented here.

Prashant Sharma • May 19 '20

@Dimitri Did you tried this -

Kindly Look at the result below -

@timeit
def append_outside_loop(limit):
    nums = []
    append = nums.append
    for num in limit:
        append(num)

append_outside_loop(list(range(1, 9999999)))

o/p - function - append_outside_loop, took 445 ms to complete

and as you said -

@timeit
def append_outside_loop(limit):
    nums = []
    append = nums.append
    [append(num) for num in limit]

append_outside_loop(list(range(1, 9999999)))

o/p - function - append_outside_loop, took 602 ms to complete

Dimitri Merejkowsky • May 19 '20

Excellent! You took the time to verify what I was saying without giving proof - and you we right to do so!

The problem is that the last line in the second function is actually building a list.

Here's a better example:

def is_even(x):
    return x % 2 == 0

@timeit
def func1(my_list):
   """ calling append() in a loop """
    my_other_list = []
    for x in my_list:
        if is_even(x):
            my_other_list.append(x)


func1(list(range(1, 9999999)))

@timeit
def func2(my_list):
    """Saving one 'op' by not looking up the is_even 
   function every time
    """
    my_other_list = []
    func = is_even
    for x in my_list:
        if func(x):
            my_other_list.append(x)


func2(list(range(1, 9999999)))

@timeit
def func3(my_list):
    """Using list comprehensions
    """
    my_other_list = [x for x in my_list if is_even(x)]

func3(list(range(1, 9999999)))

milan997 • May 4 '20

This makes me thing of how slow really Python must be if the evaluation takes so long... Anyone got more info that, how much that affects performance and why?

João Veiga • May 4 '20

Great article!
Check out the dis module and disassemble the code and it explains why this happens. TLDR: it's one less instruction inside the hot loop.