fix: avoid creating a dict and hashing boolean values

[ThomasByr]

Hi!

While reading the source code, I found that in about_time/features.py, the way conv_space is written adds unnecessary latency. While being nice, indexing and creating a dictionary each time can waste some time.

In fact, this can be measured.

def hash_stuff(value: bool):
    return {True: " ", False: ""}[value]


def return_stuff(value: bool):
    return " " if value else ""


def old_return_stuff(value: bool):
    if value:
        return " "
    else:
        return ""


def index_stuff(value: bool):
    return ["", " "][value]


cache = {True: " ", False: ""}


def hash_cached_stuff(value: bool):
    return cache[value]


if __name__ == "__main__":
    from random import randint

    from about_time import about_time

    N = 100_000_000
    Y = [bool(randint(0, 1)) for _ in range(N)]

    with about_time() as t0:
        for x in Y:
            hash_stuff(x)

    with about_time() as t1:
        for x in Y:
            return_stuff(x)

    with about_time() as t2:
        for x in Y:
            index_stuff(x)

    with about_time() as t3:
        for x in Y:
            hash_cached_stuff(x)

    with about_time() as t4:
        for x in Y:
            old_return_stuff(x)

    print(f"Hash time:         {t0.duration_human}")
    print(f"Return time:       {t1.duration_human}")
    print(f"Index time:        {t2.duration_human}")
    print(f"Cached Hash time:  {t3.duration_human}")
    print(f"Old Return time:   {t4.duration_human}")

Hash time:         12.95s
Return time:       6.2s
Index time:        9.23s
Cached Hash time:  7.03s
Old Return time:   6.13s

While ternary operations perform slightly worse than just checking (which is obscure to me), I think we can live with it to still have a one-liner.

[rsalmei]

Thanks, @ThomasByr. But I believe it’s clearer how it currently stands (perhaps not the if one, but definitely the matrix).