A Numexpr example

Numexpr is a library for the fast execution of array transformation. One can define complex elementwise operations on array and Numexpr will generate efficient code to execute the operations.

import numexpr as ne
import numpy as np

Numexpr provides fast multithreaded operations on array elements. Let’s test it on some large arrays.

a = np.random.rand(1000000)
b = np.random.rand(1000000)

Componentwise addition is easy.

ne.evaluate("a + 1")
ne.evaluate("a + b")

array([1.3195833 , 0.92546223, 1.68758307, ..., 1.19557921, 1.19559017,
       0.24145174])

We can evaluate complex expressions.

ne.evaluate('a*b-4.1*a > 2.5*b')

array([False, False, False, ..., False, False, False])

Let’s compare the performance with Numpy.

%%timeit
a * b - 4.1 * a > 2.5 * b

7.89 ms ± 91.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

%%timeit
ne.evaluate('a*b-4.1*a > 2.5*b')

995 µs ± 40.5 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

Numexpr is a factor 10 faster compared to Numpy, a nice improvement with very little effort. Let us compare some more complex operations.

%%timeit
ne.evaluate("sin(a) + arcsinh(a/b)")

5.4 ms ± 100 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

We can compare it with Numpy.

%%timeit
np.sin(a) + np.arcsinh(a / b)

39.5 ms ± 664 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)