numpy:如何加入数组？(得到几个范围的联合)

Question

我用Python numpy.

我有一个numpy索引数组a:

>>> a
array([[5, 7],
       [12, 18],
       [20, 29]])
>>> type(a)
<type 'numpy.ndarray'>

我有一个numpy索引数组b:

>>> b
array([[2, 4],
       [8, 11],
       [33, 35]])
>>> type(b)
<type 'numpy.ndarray'>

我需要加入一个数组a的数组b:

a+ b=>[2, 4] [5, 7] [8, 11] [12, 18] [20, 29] [33, 35]

=> a并且b有索引数组=> [2, 18] [20, 29] [33, 35]

(索引([2, 4][5, 7][8, 11][12, 18])按顺序进行

=> 2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18=> [2, 18])

对于这个例子:

>>> out_c
array([[2, 18],
       [20, 29],
       [33, 35]])

有人可以建议,我该怎么办out_c？

更新:@Geoff建议解决多个范围的python联合.这个解决方案是否是最快和最好的大数据阵列？

Answer 1

（新答案）使用Numpy

ranges = np.vstack((a,b))
ranges.sort(0)

# List of non-overlapping ranges
nonoverlapping = (ranges[1:,0] - ranges[:-1,1] > 1).nonzero()[0]

# Starts are 0, and all the starts not overlapped by their predecessor
starts = np.hstack(([0], nonoverlapping + 1))

# Ends are -1 and all the ends who aren't overlapped by their successor
ends = np.hstack(( nonoverlapping, [-1]))

# Result
result = np.vstack((ranges[starts, 0], ranges[ends, 1])).T

（旧答案）使用列表和集合

import numpy as np
import itertools

def ranges(s):
    """ Converts a list of integers into start, end pairs """
    for a, b in itertools.groupby(enumerate(s), lambda(x, y): y - x):
        b = list(b)
        yield b[0][1], b[-1][1]

def intersect(*args):
    """ Converts any number of numpy arrays containing start, end pairs 
        into a set of indexes """
    s = set()
    for start, end in np.vstack(args):
        s = s | set(range(start,end+1))
    return s

a = np.array([[5,7],[12, 18],[20,29]])
b = np.array([[2,4],[8,11],[33,35]])

result = np.array(list(ranges(intersect(a,b))))

参考文献

• 如何在python中找到范围重叠？
• http://docs.python.org/2/library/sets.html
• 在Python中将整数列表转换为范围
• http://docs.python.org/2/library/itertools.html#itertools.groupby