numpy.concatenate如何在列表上工作

时间:2018-07-09 13:25:39

标签: python list numpy concatenation

我试图弄清楚为什么以下代码不起作用:

import numpy as np

failList = [[[1], [2]],
           [[3, 4, 5, 6], [7]],
           [[8], [9]],
           [[10], [11, 12]],
           [[13], [14, 15, 16]]]

goodList = [[[1], [2], [3, 4, 5, 6], [7], [8]],
           [[9], [10], [11, 12], [13], [14, 15, 16]]]

goodList2 = [[[1], [2], [3, 4, 5, 6], [7], [8]],
            [[9], [10], [11, 12], [13], [14, 15, 16]],
            [[9], [10], [11, 12], [13], [14, 15, 16]]]

myLists = [failList, goodList, goodList]

for l in myLists:
    print([len(l[i]) for i in range(len(l))])
    print([len(l[i][j]) for i in range(len(l)) for j in range(len(l[i]))])
    try:
        np.concatenate(l)
        print("worked")
    except:
        print("failed")

输出为:

[2, 2, 2, 2, 2]
[1, 1, 4, 1, 1, 1, 1, 2, 1, 3]
failed
[5, 5]
[1, 1, 4, 1, 1, 1, 1, 2, 1, 3]
worked
[5, 5, 5]
[1, 1, 4, 1, 1, 1, 1, 2, 1, 3, 1, 1, 2, 1, 3]
worked

有人可以解释为什么第一个列表不能被串联,而其他人可以吗?

3 个答案:

答案 0 :(得分:2)

concatenate从每个列表元素组成一个数组,然后将它们连接到所需的轴上。如果形状不匹配,则会引发错误:

In [80]: failList = [[[1], [2]],
    ...:            [[3, 4, 5, 6], [7]],
    ...:            [[8], [9]],
    ...:            [[10], [11, 12]],
    ...:            [[13], [14, 15, 16]]]
    ...:            
In [81]: [np.array(a) for a in failList]
Out[81]: 
[array([[1],
        [2]]),
 array([list([3, 4, 5, 6]), list([7])], dtype=object),
 array([[8],
        [9]]),
 array([list([10]), list([11, 12])], dtype=object),
 array([list([13]), list([14, 15, 16])], dtype=object)]
In [82]: [np.array(a).shape for a in failList]
Out[82]: [(2, 1), (2,), (2, 1), (2,), (2,)]
In [83]: np.concatenate([np.array(a) for a in failList])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-83-c3434632bd7e> in <module>()
----> 1 np.concatenate([np.array(a) for a in failList])

ValueError: all the input arrays must have same number of dimensions

failedList中用于不同类型数组的元素,有些是2d数字的,有些是1d对象的。 concatenate无法加入。

column_stack确实有效:

In [87]: np.column_stack(failList)
Out[87]: 
array([[1, list([3, 4, 5, 6]), 8, list([10]), list([13])],
       [2, list([7]), 9, list([11, 12]), list([14, 15, 16])]],
      dtype=object)
In [88]: _.shape
Out[88]: (2, 5)

那是因为它将(2,)形状的数组重塑为(2,1)。现在,它具有一个5(2,1)数组的列表,可以在第二维上进行联接,从而生成(2,5)数组。但是请注意,它是对象dtype。有些元素是整数,有些是列表(大小不同)。

答案 1 :(得分:1)

您在串联元组(或列表)中的列表应具有相同的维度

您可以在实现np.concatenate的{​​{3}}中看到第399行。

if (PyArray_NDIM(arrays[iarrays]) != ndim) {
            PyErr_SetString(PyExc_ValueError,
                            "all the input arrays must have same "
                            "number of dimensions");
            return NULL;
}

PyArray_NDIM做到赋予所有尺寸的长度

对于您来说,failList中的列表没有相同的尺寸。 您可以通过下面的代码进行检查。

import numpy as np

failList = [[[1], [2]],
       [[3, 4, 5, 6], [7]],
       [[8], [9]],
       [[10], [11, 12]],
       [[13], [14, 15, 16]]]

goodList = [[[1], [2], [3, 4, 5, 6], [7], [8]],
[[9], [10], [11, 12], [13], [14, 15, 16]]]


goodList2 = [[[1], [2], [3, 4, 5, 6], [7], [8]],
       [[9], [10], [11, 12], [13], [14, 15, 16]],
       [[9], [10], [11, 12], [13], [14, 15, 16]]]

faileShapes = [np.shape(i) for i in failList]
print(faileShapes)

goodShapes = [np.shape(i) for i in goodList]
print(goodShapes)

goodShapes2 = [np.shape(i) for i in goodList2]
print(goodShapes2)

# printed console
# [(2, 1), (2,), (2, 1), (2,), (2,)]
# [(5,), (5,)]
# [(5,), (5,), (5,)]

答案 2 :(得分:0)

原始答案(错误):

根据docs

  

除了与轴对应的尺寸(默认为第一个)外,数组的形状必须相同。

您的第一个列表具有以下属性:内部列表的长度不同(分别为6和4)。在您的良好列表中,所有内部列表的长度都相同。5。

编辑:抱歉,我没有注意到其中一个括号,因此我错误地将您的failList的形状视为错误。

正确的答案是在failList中,子列表具有不同的形状

>>> np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]).shape
(3,3) # because all lists have the same lengths, so NumPy treats as multi-dim array

>>> np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9, 10]]).shape
(3,) # because all lists have different lengths, so NumPy treats as an array of lists