有没有比使用for循环更简洁的方法

时间:2013-04-08 07:15:54

标签: python numpy

我想知道是否有其他方法来循环和操作位于不同数组中的数据。

import numpy as np

a = np.arange(2)
b = np.arange(5)
c = np.arange(5)

l1 = []
for x in a:
    l2 = []
    for y in b: 
        l3 = []
        y = x + 1  
        for z in c:
            z = x + y
            t = (x,y,z)
            l3.append(t)
        l2.append(l3)
    l1.append(l2) 
print l1

3 个答案:

答案 0 :(得分:7)

此代码正是您正在做的事情。

def method(lst, range1, range2):
    for i in lst:
        yield [[(i, i+1, 1+(i*2))]*range2]*range1

甚至可以变成生成器表达式:

def gen_method(lst, r1, r2):
    return ([[(i, i+1, 1+(i*2))]*r2]*r1 for i in lst)

如果你愿意,可以自己测试一下。


我的测试:

a = range(2)
b = range(5)
c = range(5)

def your_method(a, b, c):
    l1 = []
    for x in a:
        l2 = []
        for y in b: 
            l3 = []
            y = x + 1  
            for z in c:
                z = x + y
                t = (x,y,z)
                l3.append(t)
            l2.append(l3)
        l1.append(l2)
    return l1

def my_method(lst, range1, range2):
    for i in lst:
        yield [[(i, i+1, 1+(i*2))]*range2]*range1

yours = your_method(a, b, c)
mine = list(my_method(a, len(b), len(c)))

print yours
print '==='
print mine
print '==='
print yours == mine

>>> 
[[[(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)]], [[(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)]]]
===
[[[(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)], [(0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1), (0, 1, 1)]], [[(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)], [(1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3), (1, 2, 3)]]]
===
True

答案 1 :(得分:1)

好吧,你可以使用列表推导来压缩代码:

[[[(x,x+1,x*2 +1)]*len(c)]*len(b) for x in a]

这样做是为a中的所有x循环,并创建元素列表,其中每个元素是为b中的所有y生成的列表,其中的每个元素列表为{{ 1}}表示c中的所有z。

答案 2 :(得分:0)

另一种方法是使用itertools,将结果列表转换为numpy array并重新整形

import numpy as np
import itertools as it

a = np.arange(2)
b = np.arange(5)
c = np.arange(5)

l1 = np.array([(i, i+1, i*2+1) for i, rb, rb in it.product(a, b, c)])
l1 = l1.reshape(len(a), len(b), len(c), len(d[0]))

随着大小的增加,这可能会比其他所有方法消耗更多内存,但只有两行,为每个三元组创建唯一元素,而不是只创建多个指针同一个对象。

修改

另一种方式,也允许将三元组(i,i + 1,i * 1 + 1)保持为列表:

l1 = np.empty([len(a), len(b), len(c)], dtype=object)
for i, rb, rb in it.product(a, b, c):
    l1[i,ra, rb] = (i, i+1, i*2+1)