为什么 numpy 数组比 python 中的列表快？

Question

I have recently discovered that numpy array is faster than normal list in python.I have run the code below

import numpy as np

import matplotlib.pyplot as plt

import pandas as pd

my_listy = list(range(1000000))

my_array = np.array(range(1000000))

%time for i in range(10): my_list2 = my_listy * 2

%time for i in range(10): my_array2 = my_array * 2

I found that execution time for the list was approx 425 ms whereas the execution time for the numpy array was approx 371 ms. Whats the specific reason for this??

Answer 1

You are performing two completely different operations here so you cannot directly compare: multiplying a list by 2 will create a new list where the list is concatenated to itself (so the length of the output list is twice that of the input list), whereas multiplying a numpy array by 2 will create a new array of the same length as the original array, but in which each element has been multiplied by 2.

Nonetheless, if you had attempted to perform a list operation which actually corresponds to the numpy case (element-by-element multiplication), for example:

my_list2 = [n * 2 for n in my_listy]

you would also have found that the numpy example was quicker. This is because the required looping in numpy is performed in a shared library consisting of compiled C code, rather than using an explicit loop in Python ( for loop or list comprehension).

$ python -mtimeit -s 'import numpy as np; my_array = np.array(range(1000000))' 'my_array2 = my_array * 2'
1000 loops, best of 3: 1.45 msec per loop

$ python -mtimeit -s 'my_listy = list(range(1000000))' 'my_list2 = [n*2 for n in my_listy]'
10 loops, best of 3: 50.8 msec per loop

Answer 2

The following are the main reasons behind the fast speed of Numpy.
-Numpy array is a collection of similar data-types that are densely packed in memory. A Python list can have different data-types, which puts lots of extra constraints while doing computation on it.
-Numpy is able to divide a task into multiple subtasks and process them parallelly.
-Numpy functions are implemented in C. Which again makes it faster compared to Python Lists.

Python at first was not made for numeric operations but with time numpy was created for this scope(making python better at numeric operations). Source: towardsdatascience.com

Answer 3

In Numpy You have implementation in C++, that this is faster than pure python.