Performant Python

Solutions¶

Ranges

First execute the code as provided

In [5]:

size = 1000000
%timeit for x in range(size): x**2
%timeit for x in np.arange(size): x**2

329 ms ± 19.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
288 ms ± 14.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

The peformance here shows little apparent benefit from using numpy. The problem is that while we are using a numpy's arange to perform the loop, the calculation x**2 is the same built-in function being called in the same fashion.

If we want to make use of numpy's vectorisation we need to simplify our code:

In [6]:

%timeit np.arange(size)**2

6.03 ms ± 775 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

We now see a significant improvement in performance. Is this accessing the billions of calculations per second that modern processors promise? Try this with other maths functions/operations.

size = 1000000
%timeit for x in range(size): x**2
%timeit for x in np.arange(size): x**2

size = 1000000
%timeit for x in range(size): x**2
%timeit for x in np.arange(size): x**2

329 ms ± 19.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
288 ms ± 14.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

The peformance here shows little apparent benefit from using numpy. The problem is that while we are using a numpy's arange to perform the loop, the calculation x**2 is the same built-in function being called in the same fashion.

If we want to make use of numpy's vectorisation we need to simplify our code:

%timeit np.arange(size)**2

%timeit np.arange(size)**2

6.03 ms ± 775 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

We now see a significant improvement in performance. Is this accessing the billions of calculations per second that modern processors promise? Try this with other maths functions/operations.