2024-heraklion-data/exercises/.ipynb_checkpoints/slow_when_slicing_wrongly-checkpoint.ipynb

In [1]:

import numpy as np

In [2]:

def get_slice_and_perform_something(x, axis):
    idx = 75
    if axis == 0:
        slice_ = x[idx, :]
    else:
        slice_ =  x[:, idx]
    # Here I divide by two but any other operation will do, 
    # we just want to simulate the fact that we actually need to read the memory
    return slice_ // 2

In [3]:

page_size = 4096
n = 100
x = np.empty((page_size * n, page_size * n), dtype='int8')

In [4]:

x.shape

Out[4]:

(409600, 409600)

In [5]:

%timeit get_slice_and_perform_something(x, axis=0)

28.2 µs ± 1.22 µs per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

In [6]:

%timeit get_slice_and_perform_something(x, axis=1)

The slowest run took 5.29 times longer than the fastest. This could mean that an intermediate result is being cached.
886 ms ± 337 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [10]:

886000 / 27.5

Out[10]:

32218.18181818182

In [ ]: