Why vectorize in Python

The same sum, three ways, and the reason numpy exists.

import timeit

import numpy as np

values = np.arange(100_000, dtype=np.float64)


def with_loop():
    total = 0.0
    for value in values:
        total += value
    return total


def with_numpy():
    return values.sum()


loop = timeit.timeit(with_loop, number=3) / 3
fast = timeit.timeit(with_numpy, number=3) / 3
print(f"loop  {loop * 1000:.1f} ms")
print(f"numpy {fast * 1000:.3f} ms")
print(f"{loop / fast:.0f}x")

How it works

1. A Python loop pays interpreter cost per element.
2. A ufunc does the loop in compiled code.
3. timeit on a small array can mislead: measure the real size.

Keywords and builtins used here

• as
• def
• for
• print
• return
• with_loop
• with_numpy

The run, in numbers

Lines

23

Characters to type

398

Tokens

122

Three-star pace

110 tpm

At the three-star pace of 110 tokens a minute, this run takes about 67 seconds.

Type this snippet

Step 4 of 4 in Maths & memory, step 19 of 22 in NumPy.

← Previous Next →