python - Numpy.argsort - 看不出有什么问题

Question

我正在尝试使用 argsort 函数对 numpy 数组进行排序。

不幸的是，这不起作用，我不明白为什么 :(

代码是:

import numpy as np

distance = np.array([38.26,  33.01,  32.33,  30.77,  37.96,  44.37,  32.72,  36.56,
        27.77,  33.62,  42.85,  34.6 ,  32.04,  27.49,  49.64,  51.85,
        44.37,  38.26,  46.93,  40.45,  40.72,  39.7 ,  34.12,  36.9 ,
        34.6 ,  34.  ,  36.56,  39.29,  38.6 ,  32.33,  32.65,  40.72,
        43.85,  47.89,  33.62,  35.24,  42.5 ,  36.97,  28.36,  37.57,
        37.25,  25.54,  29.6 ,  37.25,  40.45,  32.04,  40.45,  31.4 ,
        41.78,  35.89,  59.24,  51.2 ,  57.22,  35.54,  50.09,  40.33,
        50.58,  29.77,  51.97,  34.33,  29.  ,  43.81,  40.84,  45.62,
        39.77,  54.5 ,  40.36,  40.93,  43.28,  37.61,  45.05,  45.05,
        45.94,  45.05,  49.37,  52.56,  54.08,  53.89,  44.41,  39.25,
        36.01,  36.01,  40.93,  43.29,  38.16,  47.56,  54.5 ,  44.98,
        40.36,  36.5 ,  37.01,  46.21,  40.4 ,  30.29,  38.65,  41.49,
        40.9 ,  46.85,  32.26,  40.33,  50.58,  40.93,  59.41,  48.1 ,
        51.25,  66.76,  30.26,  61.7 ,  51.14,  64.8 ,  52.49,  48.25,
        55.24,  38.74,  41.48,  51.2 ,  51.25,  73.73,  66.05,  40.84,
        57.85,  39.2 ,  67.13,  46.98,  55.78,  62.08,  46.28,  46.21,
        48.8 ,  60.84,  62.6 ,  76.85,  48.8 ,  47.53,  43.97,  68.29,
        51.25,  50.57,  45.  ,  57.22,  54.5 ,  57.22,  40.93,  56.48,
        55.78,  53.89,  45.94,  51.25,  50.  ,  43.81])

为了调试它，我将距离数组与 argsort 一起打印:

print np.array(zip(distance, distance.argsort()))

我得到的是这个。它看起来不对，因为例如 30.77 小于 32.33，但 32.33 标记为 8，30.77 标记为 38。我做错了什么？

[[  38.26   41.  ]
 [  33.01   13.  ]
 [  32.33    8.  ]
 [  30.77   38.  ]
 [  37.96   60.  ]
 [  44.37   42.  ]
 [  32.72   57.  ]
 [  36.56  106.  ]
 [  27.77   93.  ]
 [  33.62    3.  ]
 [  42.85   47.  ]
 [  34.6    45.  ]
 [  32.04   12.  ]
 [  27.49   98.  ]
 [  49.64    2.  ]
 [  51.85   29.  ]
 [  44.37   30.  ]
 [  38.26    6.  ]
 [  46.93    1.  ]
 [  40.45    9.  ]
 [  40.72   34.  ]
 [  39.7    25.  ]
 [  34.12   22.  ]
 [  36.9    59.  ]
 [  34.6    24.  ]
 [  34.     11.  ]
 [  36.56   35.  ]
 [  39.29   53.  ]
 [  38.6    49.  ]
 [  32.33   80.  ]
 [  32.65   81.  ]
 [  40.72   89.  ]
 [  43.85   26.  ]
 [  47.89    7.  ]
 [  33.62   23.  ]
 [  35.24   37.  ]
 [  42.5    90.  ]
 [  36.97   43.  ]
 [  28.36   40.  ]
 [  37.57   39.  ]
 [  37.25   69.  ]
 [  25.54    4.  ]
 [  29.6    84.  ]
 [  37.25    0.  ]
 [  40.45   17.  ]
 [  32.04   28.  ]
 [  40.45   94.  ]
 [  31.4   113.  ]
 [  41.78  121.  ]
 [  35.89   79.  ]
 [  59.24   27.  ]
 [  51.2    21.  ]
 [  57.22   64.  ]
 [  35.54   99.  ]
 [  50.09   55.  ]
 [  40.33   88.  ]
 [  50.58   66.  ]
 [  29.77   92.  ]
 [  51.97   46.  ]
 [  34.33   44.  ]
 [  29.     19.  ]
 [  43.81   20.  ]
 [  40.84   31.  ]
 [  45.62   62.  ]
 [  39.77  119.  ]
 [  54.5    96.  ]
 [  40.36   67.  ]
 [  40.93  101.  ]
 [  43.28  142.  ]
 [  37.61   82.  ]
 [  45.05  114.  ]
 [  45.05   95.  ]
 [  45.94   48.  ]
 [  45.05   36.  ]
 [  49.37   10.  ]
 [  52.56   68.  ]
 [  54.08   83.  ]
 [  53.89  149.  ]
 [  44.41   61.  ]
 [  39.25   32.  ]
 [  36.01  134.  ]
 [  36.01    5.  ]
 [  40.93   16.  ]
 [  43.29   78.  ]
 [  38.16   87.  ]
 [  47.56  138.  ]
 [  54.5    73.  ]
 [  44.98   71.  ]
 [  40.36   70.  ]
 [  36.5    63.  ]
 [  37.01  146.  ]
 [  46.21   72.  ]
 [  40.4   127.  ]
 [  30.29   91.  ]
 [  38.65  126.  ]
 [  41.49   97.  ]
 [  40.9    18.  ]
 [  46.85  123.  ]
 [  32.26  133.  ]
 [  40.33   85.  ]
 [  50.58   33.  ]
 [  40.93  103.  ]
 [  59.41  111.  ]
 [  48.1   132.  ]
 [  51.25  128.  ]
 [  66.76   74.  ]
 [  30.26   14.  ]
 [  61.7   148.  ]
 [  51.14   54.  ]
 [  64.8   137.  ]
 [  52.49  100.  ]
 [  48.25   56.  ]
 [  55.24  108.  ]
 [  38.74  115.  ]
 [  41.48   51.  ]
 [  51.2   104.  ]
 [  51.25  136.  ]
 [  73.73  116.  ]
 [  66.05  147.  ]
 [  40.84   15.  ]
 [  57.85   58.  ]
 [  39.2   110.  ]
 [  67.13   75.  ]
 [  46.98  145.  ]
 [  55.78   77.  ]
 [  62.08   76.  ]
 [  46.28   65.  ]
 [  46.21  140.  ]
 [  48.8    86.  ]
 [  60.84  112.  ]
 [  62.6   144.  ]
 [  76.85  124.  ]
 [  48.8   143.  ]
 [  47.53  139.  ]
 [  43.97  141.  ]
 [  68.29   52.  ]
 [  51.25  120.  ]
 [  50.57   50.  ]
 [  45.    102.  ]
 [  57.22  129.  ]
 [  54.5   107.  ]
 [  57.22  125.  ]
 [  40.93  130.  ]
 [  56.48  109.  ]
 [  55.78  118.  ]
 [  53.89  105.  ]
 [  45.94  122.  ]
 [  51.25  135.  ]
 [  50.    117.  ]
 [  43.81  131.  ]]

Answer 1

distance.argsort() 返回索引数组。 ith<​​ 索引不会告诉您 distance 中第 ith<​​ 元素的排名。相反，ith<​​ 索引告诉您排序数组中的 ith<​​ 元素是 distance[i]。

换句话说，

idx = distance.argsort()
assert (distance[idx] == np.sort(distance)).all()

考虑这个小例子 from the docs :

In [236]: x = np.array([3, 1, 2])

In [237]: np.argsort(x)
Out[237]: array([1, 2, 0])

In [238]: x[np.argsort(x)]
Out[238]: array([1, 2, 3])

In [239]: x[1], x[2], x[0]
Out[239]: (1, 2, 3)

---

两次调用 argsort 确实可以得到 distance 中第 ith<​​ 元素的排名:

In [240]: np.argsort(np.argsort(x))
Out[240]: array([2, 0, 1])

了解其工作原理可以很好地测试您对 argsort 的理解。但是，调用 argsort 两次来查找排名是低效的。特别是对于较大的阵列 there are other, faster, ways找到排名。