python – How to add an extra column to a NumPy array

The Question :

330 people think this question is useful

Let’s say I have a NumPy array, a:

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

And I would like to add a column of zeros to get an array, b:

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

How can I do this easily in NumPy?

The Question Comments :

The Answer 1

196 people think this answer is useful

I think a more straightforward solution and faster to boot is to do the following:

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

And timings:

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

The Answer 2

364 people think this answer is useful

np.r_[ ... ] and np.c_[ ... ] are useful alternatives to vstack and hstack, with square brackets [] instead of round ().
A couple of examples:

: 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.])

(The reason for square brackets [] instead of round () is that Python expands e.g. 1:4 in square — the wonders of overloading.)

The Answer 3

163 people think this answer is useful

Use 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]])

The Answer 4

62 people think this answer is useful

One way, using hstack, is:

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

The Answer 5

47 people think this answer is useful

I find the following most elegant:

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

An advantage of insert is that it also allows you to insert columns (or rows) at other places inside the array. Also instead of inserting a single value you can easily insert a whole vector, for instance duplicate the last column:

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

Which leads to:

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

For the timing, insert might be slower than JoshAdel’s solution:

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

The Answer 6

46 people think this answer is useful

I was also interested in this question and compared the speed of

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

which all do the same thing for any input vector a. Timings for growing a:

enter image description here

Note that all non-contiguous variants (in particular stack/vstack) are eventually faster than all contiguous variants. column_stack (for its clarity and speed) appears to be a good option if you require contiguity.


Code to reproduce the plot:

import numpy
import perfplot

perfplot.save(
    "out.png",
    setup=lambda n: numpy.random.rand(n),
    kernels=[
        lambda a: numpy.c_[a, a],
        lambda a: numpy.ascontiguousarray(numpy.stack([a, a]).T),
        lambda a: numpy.ascontiguousarray(numpy.vstack([a, a]).T),
        lambda a: numpy.column_stack([a, a]),
        lambda a: numpy.concatenate([a[:, None], a[:, None]], axis=1),
        lambda a: numpy.ascontiguousarray(
            numpy.concatenate([a[None], a[None]], axis=0).T
        ),
        lambda a: numpy.stack([a, a]).T,
        lambda a: numpy.vstack([a, a]).T,
        lambda a: numpy.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(20)],
    xlabel="len(a)",
    logx=True,
    logy=True,
)

The Answer 7

32 people think this answer is useful

I think:

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

is more elegant.

The Answer 8

12 people think this answer is useful

np.concatenate also works

>>> 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.]])

The Answer 9

12 people think this answer is useful

Assuming M is a (100,3) ndarray and y is a (100,) ndarray append can be used as follows:

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

The trick is to use

y[:, None]

This converts y to a (100, 1) 2D array.

M.shape

now gives

(100, 4)

The Answer 10

8 people think this answer is useful

I like JoshAdel’s answer because of the focus on performance. A minor performance improvement is to avoid the overhead of initializing with zeros, only to be overwritten. This has a measurable difference when N is large, empty is used instead of zeros, and the column of zeros is written as a separate step:

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

The Answer 11

8 people think this answer is useful

np.insert also serves the purpose.

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]])

It inserts values, here new_col, before a given index, here idx along one axis. In other words, the newly inserted values will occupy the idx column and move what were originally there at and after idx backward.

The Answer 12

7 people think this answer is useful

Add an extra column to a numpy array:

Numpy’s np.append method takes three parameters, the first two are 2D numpy arrays and the 3rd is an axis parameter instructing along which axis to append:

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)) 

Prints:

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

The Answer 13

4 people think this answer is useful

A bit late to the party, but nobody posted this answer yet, so for the sake of completeness: you can do this with list comprehensions, on a plain Python array:

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

The Answer 14

4 people think this answer is useful

For me, the next way looks pretty intuitive and simple.

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

The Answer 15

3 people think this answer is useful

In my case, I had to add a column of ones to a NumPy array

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)

After X.shape => (97, 2)

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

The Answer 16

1 people think this answer is useful

There is a function specifically for this. It is called 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]])

Here is what it says in the docstring:

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]])

The Answer 17

1 people think this answer is useful

I liked this:

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

Add a Comment