如何向 NumPy 数组添加额外的列？

问题描述：

给定以下二维数组：

a = np.array([
    [1, 2, 3],
    [2, 3, 4],
])

我想沿第二轴添加一列零以获得：

b = np.array([
    [1, 2, 3, 0],
    [2, 3, 4, 0],
])

解决方案 1：

np.r_[...]( docs ) 和np.c_[...]( docsnp.vstack ) 是and的有用替代np.hstack。请注意，它们使用方括号 [] 而不是圆括号 ()。

一些例子：

: import numpy as np
: N = 3
: A = np.eye(N)

: np.c_[ A, np.ones(N) ]              # add a column
array([[ 1.,  0.,  0.,  1.],
       [ 0.,  1.,  0.,  1.],
       [ 0.,  0.,  1.,  1.]])

: np.c_[ np.ones(N), A, np.ones(N) ]  # or two
array([[ 1.,  1.,  0.,  0.,  1.],
       [ 1.,  0.,  1.,  0.,  1.],
       [ 1.,  0.,  0.,  1.,  1.]])

: np.r_[ A, [A[1]] ]              # add a row
array([[ 1.,  0.,  0.],
       [ 0.,  1.,  0.],
       [ 0.,  0.,  1.],
       [ 0.,  1.,  0.]])
: # not np.r_[ A, A[1] ]

: np.r_[ A[0], 1, 2, 3, A[1] ]    # mix vecs and scalars
  array([ 1.,  0.,  0.,  1.,  2.,  3.,  0.,  1.,  0.])

: np.r_[ A[0], [1, 2, 3], A[1] ]  # lists
  array([ 1.,  0.,  0.,  1.,  2.,  3.,  0.,  1.,  0.])

: np.r_[ A[0], (1, 2, 3), A[1] ]  # tuples
  array([ 1.,  0.,  0.,  1.,  2.,  3.,  0.,  1.,  0.])

: np.r_[ A[0], 1:4, A[1] ]        # same, 1:4 == arange(1,4) == 1,2,3
  array([ 1.,  0.,  0.,  1.,  2.,  3.,  0.,  1.,  0.])

使用方括号 [ ] 而不是圆括号 () 的原因是 Python 会转换1:4为方括号中的切片对象。

解决方案 2：

我认为更直接、启动速度更快的解决方案是执行以下操作：

import numpy as np
N = 10
a = np.random.rand(N,N)
b = np.zeros((N,N+1))
b[:,:-1] = a

时间安排如下：

In [23]: N = 10

In [24]: a = np.random.rand(N,N)

In [25]: %timeit b = np.hstack((a,np.zeros((a.shape[0],1))))
10000 loops, best of 3: 19.6 us per loop

In [27]: %timeit b = np.zeros((a.shape[0],a.shape[1]+1)); b[:,:-1] = a
100000 loops, best of 3: 5.62 us per loop

解决方案 3：

使用numpy.append：

>>> a = np.array([[1,2,3],[2,3,4]])
>>> a
array([[1, 2, 3],
       [2, 3, 4]])

>>> z = np.zeros((2,1), dtype=int64)
>>> z
array([[0],
       [0]])

>>> np.append(a, z, axis=1)
array([[1, 2, 3, 0],
       [2, 3, 4, 0]])

解决方案 4：

其中一种方法是使用hstack，其方法是：

b = np.hstack((a, np.zeros((a.shape[0], 1), dtype=a.dtype)))

解决方案 5：

我也对这个问题感兴趣，并比较了

numpy.c_[a, a]
numpy.stack([a, a]).T
numpy.vstack([a, a]).T
numpy.ascontiguousarray(numpy.stack([a, a]).T)               
numpy.ascontiguousarray(numpy.vstack([a, a]).T)
numpy.column_stack([a, a])
numpy.concatenate([a[:,None], a[:,None]], axis=1)
numpy.concatenate([a[None], a[None]], axis=0).T

它们对任何输入向量都做同样的事情a。增长时间a：

请注意，所有非连续变体（特别是 stack/ vstack）最终都比所有连续变体更快。column_stack如果您需要连续性，（因为它的清晰度和速度）似乎是一个不错的选择。

重现情节的代码：

import numpy as np
import perfplot

b = perfplot.bench(
    setup=np.random.rand,
    kernels=[
        lambda a: np.c_[a, a],
        lambda a: np.ascontiguousarray(np.stack([a, a]).T),
        lambda a: np.ascontiguousarray(np.vstack([a, a]).T),
        lambda a: np.column_stack([a, a]),
        lambda a: np.concatenate([a[:, None], a[:, None]], axis=1),
        lambda a: np.ascontiguousarray(np.concatenate([a[None], a[None]], axis=0).T),
        lambda a: np.stack([a, a]).T,
        lambda a: np.vstack([a, a]).T,
        lambda a: np.concatenate([a[None], a[None]], axis=0).T,
    ],
    labels=[
        "c_",
        "ascont(stack)",
        "ascont(vstack)",
        "column_stack",
        "concat",
        "ascont(concat)",
        "stack (non-cont)",
        "vstack (non-cont)",
        "concat (non-cont)",
    ],
    n_range=[2 ** k for k in range(23)],
    xlabel="len(a)",
)
b.save("out.png")

解决方案 6：

我发现以下最优雅：

b = np.insert(a, 3, values=0, axis=1) # Insert values before column 3

其优点insert是它还允许您在数组内的其他位置插入列（或行）。此外，您可以轻松地插入整个向量，而不是插入单个值，例如复制最后一列：

b = np.insert(a, insert_index, values=a[:,2], axis=1)

这导致：

array([[1, 2, 3, 3],
       [2, 3, 4, 4]])

就时间而言，insert可能比 JoshAdel 的解决方案慢：

In [1]: N = 10

In [2]: a = np.random.rand(N,N)

In [3]: %timeit b = np.hstack((a, np.zeros((a.shape[0], 1))))
100000 loops, best of 3: 7.5 µs per loop

In [4]: %timeit b = np.zeros((a.shape[0], a.shape[1]+1)); b[:,:-1] = a
100000 loops, best of 3: 2.17 µs per loop

In [5]: %timeit b = np.insert(a, 3, values=0, axis=1)
100000 loops, best of 3: 10.2 µs per loop

解决方案 7：

我认为：

np.column_stack((a, zeros(shape(a)[0])))

更加优雅。

解决方案 8：

假设M是一个 (100,3) ndarray 并且y是一个 (100,) ndarray，append可以按如下方式使用：

M=numpy.append(M,y[:,None],1)

诀窍是使用

y[:, None]

这将转换y为 (100, 1) 二维数组。

M.shape

现在给出

(100, 4)

解决方案 9：

向 numpy 数组添加一个额外的列：

Numpy 的np.append方法需要三个参数，前两个是二维 numpy 数组，第三个是轴参数，指示沿哪个轴附加：

import numpy as np  
x = np.array([[1,2,3], [4,5,6]]) 
print("Original x:") 
print(x) 

y = np.array([[1], [1]]) 
print("Original y:") 
print(y) 

print("x appended to y on axis of 1:") 
print(np.append(x, y, axis=1)) 

印刷：

Original x:
[[1 2 3]
 [4 5 6]]
Original y:
[[1]
 [1]]
y appended to x on axis of 1:
[[1 2 3 1]
 [4 5 6 1]]

解决方案 10：

np.concatenate也有效

>>> a = np.array([[1,2,3],[2,3,4]])
>>> a
array([[1, 2, 3],
       [2, 3, 4]])
>>> z = np.zeros((2,1))
>>> z
array([[ 0.],
       [ 0.]])
>>> np.concatenate((a, z), axis=1)
array([[ 1.,  2.,  3.,  0.],
       [ 2.,  3.,  4.,  0.]])

解决方案 11：

np.insert也能达到目的。

matA = np.array([[1,2,3], 
                 [2,3,4]])
idx = 3
new_col = np.array([0, 0])
np.insert(matA, idx, new_col, axis=1)

array([[1, 2, 3, 0],
       [2, 3, 4, 0]])

new_col它会沿一个轴在给定索引（此处）之前插入值（此处）idx。换句话说，新插入的值将占据该idx列，并将原来位于该列及其之后的值idx向后移动。

解决方案 12：

我喜欢 JoshAdel 的回答，因为他注重性能。一个小的性能改进是避免用零初始化的开销，而只需覆盖即可。当 N 很大、使用空而不是零并且将零列写为单独的步骤时，这有可测量的差异：

In [1]: import numpy as np

In [2]: N = 10000

In [3]: a = np.ones((N,N))

In [4]: %timeit b = np.zeros((a.shape[0],a.shape[1]+1)); b[:,:-1] = a
1 loops, best of 3: 492 ms per loop

In [5]: %timeit b = np.empty((a.shape[0],a.shape[1]+1)); b[:,:-1] = a; b[:,-1] = np.zeros((a.shape[0],))
1 loops, best of 3: 407 ms per loop

解决方案 13：

对我来说，下一种方法看起来非常直观和简单。

zeros = np.zeros((2,1)) #2 is a number of rows in your array.   
b = np.hstack((a, zeros))

解决方案 14：

有点晚了，但是还没有人发布这个答案，所以为了完整起见：你可以在普通的 Python 数组上使用列表推导来做到这一点：

source = a.tolist()
result = [row + [0] for row in source]
b = np.array(result)

解决方案 15：

在我的例子中，我必须向 NumPy 数组中添加一列

X = array([ 6.1101, 5.5277, ... ])
X.shape => (97,)
X = np.concatenate((np.ones((m,1), dtype=np.int), X.reshape(m,1)), axis=1)

X.shape => (97, 2) 之后

array([[ 1. , 6.1101],
       [ 1. , 5.5277],
...

解决方案 16：

有一个专门用于此的函数。它被称为 numpy.pad

a = np.array([[1,2,3], [2,3,4]])
b = np.pad(a, ((0, 0), (0, 1)), mode='constant', constant_values=0)
print b
>>> array([[1, 2, 3, 0],
           [2, 3, 4, 0]])

以下是文档字符串中的内容：

Pads an array.

Parameters
----------
array : array_like of rank N
    Input array
pad_width : {sequence, array_like, int}
    Number of values padded to the edges of each axis.
    ((before_1, after_1), ... (before_N, after_N)) unique pad widths
    for each axis.
    ((before, after),) yields same before and after pad for each axis.
    (pad,) or int is a shortcut for before = after = pad width for all
    axes.
mode : str or function
    One of the following string values or a user supplied function.

    'constant'
        Pads with a constant value.
    'edge'
        Pads with the edge values of array.
    'linear_ramp'
        Pads with the linear ramp between end_value and the
        array edge value.
    'maximum'
        Pads with the maximum value of all or part of the
        vector along each axis.
    'mean'
        Pads with the mean value of all or part of the
        vector along each axis.
    'median'
        Pads with the median value of all or part of the
        vector along each axis.
    'minimum'
        Pads with the minimum value of all or part of the
        vector along each axis.
    'reflect'
        Pads with the reflection of the vector mirrored on
        the first and last values of the vector along each
        axis.
    'symmetric'
        Pads with the reflection of the vector mirrored
        along the edge of the array.
    'wrap'
        Pads with the wrap of the vector along the axis.
        The first values are used to pad the end and the
        end values are used to pad the beginning.
    <function>
        Padding function, see Notes.
stat_length : sequence or int, optional
    Used in 'maximum', 'mean', 'median', and 'minimum'.  Number of
    values at edge of each axis used to calculate the statistic value.

    ((before_1, after_1), ... (before_N, after_N)) unique statistic
    lengths for each axis.

    ((before, after),) yields same before and after statistic lengths
    for each axis.

    (stat_length,) or int is a shortcut for before = after = statistic
    length for all axes.

    Default is ``None``, to use the entire axis.
constant_values : sequence or int, optional
    Used in 'constant'.  The values to set the padded values for each
    axis.

    ((before_1, after_1), ... (before_N, after_N)) unique pad constants
    for each axis.

    ((before, after),) yields same before and after constants for each
    axis.

    (constant,) or int is a shortcut for before = after = constant for
    all axes.

    Default is 0.
end_values : sequence or int, optional
    Used in 'linear_ramp'.  The values used for the ending value of the
    linear_ramp and that will form the edge of the padded array.

    ((before_1, after_1), ... (before_N, after_N)) unique end values
    for each axis.

    ((before, after),) yields same before and after end values for each
    axis.

    (constant,) or int is a shortcut for before = after = end value for
    all axes.

    Default is 0.
reflect_type : {'even', 'odd'}, optional
    Used in 'reflect', and 'symmetric'.  The 'even' style is the
    default with an unaltered reflection around the edge value.  For
    the 'odd' style, the extented part of the array is created by
    subtracting the reflected values from two times the edge value.

Returns
-------
pad : ndarray
    Padded array of rank equal to `array` with shape increased
    according to `pad_width`.

Notes
-----
.. versionadded:: 1.7.0

For an array with rank greater than 1, some of the padding of later
axes is calculated from padding of previous axes.  This is easiest to
think about with a rank 2 array where the corners of the padded array
are calculated by using padded values from the first axis.

The padding function, if used, should return a rank 1 array equal in
length to the vector argument with padded values replaced. It has the
following signature::

    padding_func(vector, iaxis_pad_width, iaxis, kwargs)

where

    vector : ndarray
        A rank 1 array already padded with zeros.  Padded values are
        vector[:pad_tuple[0]] and vector[-pad_tuple[1]:].
    iaxis_pad_width : tuple
        A 2-tuple of ints, iaxis_pad_width[0] represents the number of
        values padded at the beginning of vector where
        iaxis_pad_width[1] represents the number of values padded at
        the end of vector.
    iaxis : int
        The axis currently being calculated.
    kwargs : dict
        Any keyword arguments the function requires.

Examples
--------
>>> a = [1, 2, 3, 4, 5]
>>> np.pad(a, (2,3), 'constant', constant_values=(4, 6))
array([4, 4, 1, 2, 3, 4, 5, 6, 6, 6])

>>> np.pad(a, (2, 3), 'edge')
array([1, 1, 1, 2, 3, 4, 5, 5, 5, 5])

>>> np.pad(a, (2, 3), 'linear_ramp', end_values=(5, -4))
array([ 5,  3,  1,  2,  3,  4,  5,  2, -1, -4])

>>> np.pad(a, (2,), 'maximum')
array([5, 5, 1, 2, 3, 4, 5, 5, 5])

>>> np.pad(a, (2,), 'mean')
array([3, 3, 1, 2, 3, 4, 5, 3, 3])

>>> np.pad(a, (2,), 'median')
array([3, 3, 1, 2, 3, 4, 5, 3, 3])

>>> a = [[1, 2], [3, 4]]
>>> np.pad(a, ((3, 2), (2, 3)), 'minimum')
array([[1, 1, 1, 2, 1, 1, 1],
       [1, 1, 1, 2, 1, 1, 1],
       [1, 1, 1, 2, 1, 1, 1],
       [1, 1, 1, 2, 1, 1, 1],
       [3, 3, 3, 4, 3, 3, 3],
       [1, 1, 1, 2, 1, 1, 1],
       [1, 1, 1, 2, 1, 1, 1]])

>>> a = [1, 2, 3, 4, 5]
>>> np.pad(a, (2, 3), 'reflect')
array([3, 2, 1, 2, 3, 4, 5, 4, 3, 2])

>>> np.pad(a, (2, 3), 'reflect', reflect_type='odd')
array([-1,  0,  1,  2,  3,  4,  5,  6,  7,  8])

>>> np.pad(a, (2, 3), 'symmetric')
array([2, 1, 1, 2, 3, 4, 5, 5, 4, 3])

>>> np.pad(a, (2, 3), 'symmetric', reflect_type='odd')
array([0, 1, 1, 2, 3, 4, 5, 5, 6, 7])

>>> np.pad(a, (2, 3), 'wrap')
array([4, 5, 1, 2, 3, 4, 5, 1, 2, 3])

>>> def pad_with(vector, pad_width, iaxis, kwargs):
...     pad_value = kwargs.get('padder', 10)
...     vector[:pad_width[0]] = pad_value
...     vector[-pad_width[1]:] = pad_value
...     return vector
>>> a = np.arange(6)
>>> a = a.reshape((2, 3))
>>> np.pad(a, 2, pad_with)
array([[10, 10, 10, 10, 10, 10, 10],
       [10, 10, 10, 10, 10, 10, 10],
       [10, 10,  0,  1,  2, 10, 10],
       [10, 10,  3,  4,  5, 10, 10],
       [10, 10, 10, 10, 10, 10, 10],
       [10, 10, 10, 10, 10, 10, 10]])
>>> np.pad(a, 2, pad_with, padder=100)
array([[100, 100, 100, 100, 100, 100, 100],
       [100, 100, 100, 100, 100, 100, 100],
       [100, 100,   0,   1,   2, 100, 100],
       [100, 100,   3,   4,   5, 100, 100],
       [100, 100, 100, 100, 100, 100, 100],
       [100, 100, 100, 100, 100, 100, 100]])

解决方案 17：

我喜欢这个：

new_column = np.zeros((len(a), 1))
b = np.block([a, new_column])