I have a two-dimensional array with shape (x, y) which I want to convert to a three-dimensional array with shape (x, y, 1). Is there a nice Pythonic way to do this?
Add a comment
|
10 Answers
In addition to the other answers, you can also use slicing with numpy.newaxis:
>>> from numpy import zeros, newaxis
>>> a = zeros((6, 8))
>>> a.shape
(6, 8)
>>> b = a[:, :, newaxis]
>>> b.shape
(6, 8, 1)
Or even this (which will work with an arbitrary number of dimensions):
>>> b = a[..., newaxis]
>>> b.shape
(6, 8, 1)
8 Comments
Joe Kington
On a side note,
numpy.newaxis is just None. newaxis is "just" for readibility. It's equivalent to just do b = a[..., None] (The ellipsis also allows it to work for N-dimensional arrays, not just 2D arrays.)
Mark Dickinson
True. For some reason I had the impression that
newaxis being None was just an implementation detail (and therefore possibly subject to change in the future), but it looks like it's explicitly documented.
Gathide
Suppose you want the third axis to be something other than 1? e.g. how do you convert
a to be b with b.shape = (6,8,3)?Mark Dickinson
@Gathide: That depends. Your old array has
6*8 = 48 numbers in it. The new one has 6*8*3 = 144 numbers. How do you want to map the original 48 numbers into your new array? Repeat along the third axis? In which case, most of the time you don't bother with the repetition: leave it as shape (6, 8, 1) and make use of NumPy's ability to broadcast. But if you really need shape (6, 8, 3), look into np.tile and np.broadcast_to.
wingr
@Gathide as Mark Dickinson says you can do this with
np.broadcast_to like b = np.broadcast_to(a[..., np.newaxis], (6, 8, 3)). |
numpy.reshape(array, array.shape + (1,))
6 Comments
nobody
Thanks, I used A = A.reshape(A.shape + (1,))
Mark Dickinson
If you're happy to modify
A in place, you can simply assign to the shape attribute: A.shape = A.shape + (1,), or even A.shape += 1,.
seralouk
assuming that i have a 3D [1000,10,5] array and I want to convert it to a 2D with shape [1000, 50] like concatenating the 2nd and 3rd initial dimensions. Is
reshape fine in this case?
Devid
yea it's sure
Winston Ewert
@serafeim, sure.
seralouk
Thanks. I noticed that the desired outcome is when I use
order='F' inside np.reshape. If order='F' is NOT specified, then the output is wrong in my caseimport numpy as np
# Create a two-dimensional array
a = np.array([[1,2,3], [4,5,6], [1,2,3], [4,5,6],[1,2,3], [4,5,6],[1,2,3], [4,5,6]])
print(a.shape)
# Shape of a = (8,3)
b = np.reshape(a, (8, 3, -1))
# Changing the shape. -1 means any number which is suitable
print(b.shape)
# Size of b = (8,3,1)
Comments
import numpy as np
a= np.eye(3)
print a.shape
b = a.reshape(3,3,1)
print b.shape
Comments
I hope this function helps you to convert the two-dimensional array to a three-dimensional array.
Args:
x: 2darray, (n_time, n_in)
agg_num: int, number of frames to concatenate.
hop: int, number of hop frames.
Returns:
3darray, (n_blocks, agg_num, n_in)
def d_2d_to_3d(x, agg_num, hop):
# Pad to at least one block.
len_x, n_in = x.shape
if (len_x < agg_num): #not in get_matrix_data
x = np.concatenate((x, np.zeros((agg_num - len_x, n_in))))
# main 2d to 3d.
len_x = len(x)
i1 = 0
x3d = []
while (i1 + agg_num <= len_x):
x3d.append(x[i1 : i1 + agg_num])
i1 += hop
return np.array(x3d)
Comments
If you just want to add a third axis (x,y) to (x,y,1), NumPy allows you to easily do this using the dstack command.
import numpy as np
a = np.eye(3) # your matrix here
b = np.dstack(a).T
You need to transpose (.T) it to get it into the (x,y,1) format you want.
Comments
import numpy as np
# create a 2-D ndarray
a = np.array([[2,3,4], [5,6,7]])
print(a.ndim)
>> 2
print(a.shape)
>> (2, 3)
# add 3rd dimension
1st option: reshape
b = np.reshape(a, a.shape + (1,))
print(b.ndim)
>> 3
print(b.shape)
>> (2, 3, 1)
2nd option: expand_dims
c = np.expand_dims(a, axis=2)
print(c.ndim)
>> 3
print(c.shape)
>> (2, 3, 1)
Comments
A simple way, with some math
At first, you know the number of array elements, let’s say 100 and then divide 100 in three steps like:
25 * 2 * 2 = 100
or: 4 * 5 * 5 = 100
import numpy as np
D = np.arange(100)
# Change to three-dimensional by division of 100 for 3 steps 100 = 25 * 2 * 2
D3 = D.reshape(2, 2, 25) # 25*2*2 = 100
Another way:
another_3D = D.reshape(4, 5, 5)
print(another_3D.ndim)
To four-dimensional:
D4 = D.reshape(2, 2, 5, 5)
print(D4.ndim)
Comments
If you have an array
a2 = np.array([[1, 2, 3.3],
[4, 5, 6.5]])
then you can change this array into a three-dimensional array of shape (2, 3, 3) by using:
a2_new = np.reshape(a2, a2.shape + (1,)) a2_new
Your output will be:
array([[[1. ],
[2. ],
[3.3]],
[[4. ],
[5. ],
[6.5]]])
Or
You can try:
a2.reshape(2, 3, 1)
This will change your two-dimensional array to three-dimensional of shape(2, 3, 1).
