Home Original page

bpo-34210: Small improvements in heapq (refactoring) by Amper · Pull Request #8439 · python/cpython

@rhettinger I'm sorry, maybe my mistake is that I sent too many things in one commit.
And I agree, paragraph 2 degrades readability in favor of speed. And code formatting was superfluous. I'll take these mistakes into account in the future.

But still the main change that I wanted to focus on in paragraph 1. It's about code:

if n >= size:
    return sorted(iterable, key=key)[:n]

In my opinion, using a slice here is unnecessary (because the number of elements is always known to be no greater than n) and it makes the code slower.

According to my measurements this branch in the code faster by 12-15% if you remove the slice using:

def nlargest(n, iterable, key=None):
    ...
    try:
        size = len(iterable)
    except (TypeError, AttributeError):
        pass
    else:
        if n >= size:
            return sorted(iterable, key=key, reverse=True)[:n]
    ...

def nlargest_new(n, iterable, key=None):
    ...
    try:
        size = len(iterable)
    except (TypeError, AttributeError):
        pass
    else:
        if n >= size:
            return sorted(iterable, key=key, reverse=True)
    ...

if __name__ == "__main__":

    ...

    from timeit import timeit
    from random import randrange

    data = [randrange(10000) for i in range(10000)]

    test = """
for i in (10000, 5000, 1000, 500, 100, 10, 5, 1):
    data = data[:i]
    for n in (i, i + 1, i + 10, i + 100, i + 1000, i + 5000, i + 10000):
        result = list(nlargest(n, data))
"""

    print('before:', timeit(test, setup="from __main__ import data, nlargest", number=100000))
    print('after:', timeit(test, setup="from __main__ import data, nlargest_new as nlargest", number=100000))

Here are the results of one of the launches:

before: 5.075591027
after: 4.450232958999999

I apologize for your time.