Suppose I have a n × m array, i.e.:
array([[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]])
And I what to generate a 3D array k × n × m, where all the arrays in the new axis are equal, i.e.: the same array but now 3 × 3 × 3.
array([[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]],
[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]],
[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]]])
How can I get it?
Introduce a new axis at the start with None/np.newaxis and replicate along it with np.repeat. This should work for extending any n dim array to n+1 dim array. The implementation would be -
np.repeat(arr[None,...],k,axis=0)
Sample run -
In [143]: arr
Out[143]:
array([[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]])
In [144]: np.repeat(arr[None,...],3,axis=0)
Out[144]:
array([[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]],
[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]],
[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]]])
View-output for memory-efficiency
We can also generate a 3D view and achieve virtually free runtime with np.broadcast_to. More info - here. Hence, simply do -
np.broadcast_to(arr,(3,)+arr.shape) # repeat 3 times
if you have:
a = [[1, 2, 3], [4, 5, 6], [7, 8, 9]]
You can use a list comprehension to generate the duplicate array:
b = [a for x in range(3)]
Then (for numpy):
c = array(b)
One possibility would be to use default broadcasting to replicate your array:
a = np.arange(1, 10).reshape(3,3)
n = 3
b = np.ones((n, 3, 3)) * a
Which results in the array you wanted:
array([[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]],
[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]],
[[ 1., 2., 3.],
[ 4., 5., 6.],
[ 7., 8., 9.]]])
This won't work by default if you want to replicate it along another axis. In that case you would need to be explicit with the dimensions to ensure correct broadcasting.
I think this answer is exactly the answer of Divakar, but the syntax might be a bit easier to understand for a beginner(at least in my case, it is):
a = np.array([[1,2,3],[4,5,6]])
a[np.newaxis,:,:].repeat(3,axis=0)
results in:
array([[[1, 2, 3],
[4, 5, 6]],
[[1, 2, 3],
[4, 5, 6]],
[[1, 2, 3],
[4, 5, 6]]])
I learned about np.newaxis here: What is numpy.newaxis and when to use it.
And about numpy.repeat here: numpy.repeat
Here's an example usage I needed this for:
k = np.array([[[111,121,131,141,151],[211,221,231,241,251]],\
[[112,122,132,142,152],[212,222,232,242,252]],\
[[113,123,133,143,153],[213,223,233,243,253]]])
filter = np.array([[True,True,True,True,False],
[True,False,False,True,False]])
k[filter[None,...].repeat(3,axis=0)] = 0
print(k)
results in:
[[[ 0 0 0 0 151]
[ 0 221 231 0 251]]
[[ 0 0 0 0 152]
[ 0 222 232 0 252]]
[[ 0 0 0 0 153]
[ 0 223 233 0 253]]]
