I have a large number of 3d numpy arrays, which when assembled together, form a single contiguous 3d dataset*. However, the arrays were created by breaking the larger space into chunks. I need to assemble the chunk arrays back together. To simplify the problem, I've reduced it to the following example, with four chunks, each of which has 2x2x2 values.
So I have:
yellow_chunk = np.array([[[1,2], [5,6]], [[17,18], [21,22]]])
green_chunk = np.array([[[3,4], [7,8]], [[19,20], [23,24]]])
blue_chunk = np.array([[[9,10], [13,14]], [[25,26], [29,30]]])
red_chunk = np.array([[[11,12], [15,16]], [[27,28], [31,32]]])
And I want to end up with:
>>> output
array([[[ 1, 2, 3, 4],
[ 5, 6, 7, 8],
[ 9, 10, 11, 12],
[13, 14, 15, 16]],
[[17, 18, 19, 20],
[21, 22, 23, 24],
[25, 26, 27, 28],
[29, 30, 31, 32]]])
Illustration for this small example:
Things I've tried
concatenate
>>> np.concatenate([yellow_chunk,green_chunk,blue_chunk,red_chunk],-1)
array([[[ 1, 2, 3, 4, 9, 10, 11, 12],
[ 5, 6, 7, 8, 13, 14, 15, 16]],
[[17, 18, 19, 20, 25, 26, 27, 28],
[21, 22, 23, 24, 29, 30, 31, 32]]])
This was close, but the shape is wrong: 8x2x2 instead of the 4x2x4 I need.
hstack
>>> np.hstack([yellow_chunk,green_chunk,blue_chunk,red_chunk])
array([[[ 1, 2],
[ 5, 6],
[ 3, 4],
[ 7, 8],
[ 9, 10],
[13, 14],
[11, 12],
[15, 16]],
[[17, 18],
[21, 22],
[19, 20],
[23, 24],
[25, 26],
[29, 30],
[27, 28],
[31, 32]]])
Also the wrong shape.
vstack
>>> np.vstack([yellow_chunk,green_chunk,blue_chunk,red_chunk])
array([[[ 1, 2],
[ 5, 6]],
[[17, 18],
[21, 22]],
[[ 3, 4],
[ 7, 8]],
[[19, 20],
[23, 24]],
[[ 9, 10],
[13, 14]],
[[25, 26],
[29, 30]],
[[11, 12],
[15, 16]],
[[27, 28],
[31, 32]]])
Wrong shape and order.
dstack
>>> np.dstack([yellow_chunk,green_chunk,blue_chunk,red_chunk])
array([[[ 1, 2, 3, 4, 9, 10, 11, 12],
[ 5, 6, 7, 8, 13, 14, 15, 16]],
[[17, 18, 19, 20, 25, 26, 27, 28],
[21, 22, 23, 24, 29, 30, 31, 32]]])
Wrong shape and order.
* In reality, I have 16x16 chunks, each of which has a shape of 16x128x16. So I'm stitching together "rows" of 256 values rather than the 4-value rows that I have in my small example above.
np.block([[yellow_chunk, green_chunk], [blue_chunk, red_chunk]])
>>>
[[[ 1 2 3 4]
[ 5 6 7 8]
[ 9 10 11 12]
[13 14 15 16]]
[[17 18 19 20]
[21 22 23 24]
[25 26 27 28]
[29 30 31 32]]]
What you are doing here is assembling an nd-array from nested lists of blocks.
If you want more information about joining arrays, you can read this numpy.org doc on all the relevant methods and functions useable.
Simply this for example:
np.hstack((np.dstack((y,g)), np.dstack((b,r))))
(renaming yellow_chunk to y and so on)
Related
With given 2D and 1D lists, I have to dot product them. But I have to calculate them without using .dot.
For example, I want to make these lists
matrix_A = [[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15], [16, 17, 18, 19], [20, 21, 22, 23], [24, 25, 26, 27], [28, 29, 30, 31]]
vector_x = [0, 1, 2, 3]
to this output
result_list = [ 14 38 62 86 110 134 158 182]
How can I do it by only using lists(not using NumPy array and .dot) in python?
You could use a list comprehension with nested for loops.
matrix_A = [[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15], [16, 17, 18, 19], [20, 21, 22, 23], [24, 25, 26, 27], [28, 29, 30, 31]]
vector_x = [0, 1, 2, 3]
result_list = [sum(a*b for a,b in zip(row, vector_x)) for row in matrix_A]
print(result_list)
Output:
[14, 38, 62, 86, 110, 134, 158, 182]
Edit: Removed the square brackets in the list comprehension following #fshabashev's comment.
If you do not mind using numpy, this is a solution
import numpy as np
matrix_A = [[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11], [12, 13, 14, 15], [16, 17, 18, 19], [20, 21, 22, 23], [24, 25, 26, 27], [28, 29, 30, 31]]
vector_x = [0, 1, 2, 3]
res = np.sum(np.array(matrix_A) * np.array(vector_x), axis=1)
print(res)
I have a multidimensional array of shape (n,x,y). For this example can use this array
A = array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8],
[ 9, 10, 11]],
[[12, 13, 14],
[15, 16, 17],
[18, 19, 20],
[21, 22, 23]],
[[24, 25, 26],
[27, 28, 29],
[30, 31, 32],
[33, 34, 35]]])
I then have another multidimensional array that has index values that I want to use on the original array, A. This has shape (z,2) and the values represent row values index’s
Row_values = array([[0,1],
[0,2],
[1,2],
[1,3]])
So I want to use all the index values in row_values to apply to each of the three arrays in A so I end up with a final array of shape (12,2,3)
Result = ([[[0,1,2],
[3,4,5]],
[[0,1,2],
[6,7,8]],
[[3,4,5],
[6,7,8]]
[[3,4,5],
[9,10,11],
[[12,13,14],
[15,16,17]],
[[12,13,14],
[18,19,20]],
[[15,16,17],
[18,19,20]],
[[15,16,17],
[21,22,23]],
[[24,25,26],
[27,28,29]],
[[24,25,26],
[30,31,32]],
[[27,28,29],
[30,31,32]],
[[27,28,29],
[33,34,35]]]
I have tried using np.take() but haven’t been able to make it work. Not sure if there’s another numpy function that is easier to use
We can advantage of NumPy's advanced indexing and using np.repeat and np.tile along with it.
cidx = np.tile(Row_values, (A.shape[0], 1))
ridx = np.repeat(np.arange(A.shape[0]), Row_values.shape[0])
out = A[ridx[:, None], cidx]
# out.shape -> (12, 2, 3)
Using np.take
np.take(A, Row_values, axis=1).reshape((-1, 2, 3))
# Or
A[:, Row_values].reshape((-1, 2, 3))
Output:
array([[[ 0, 1, 2],
[ 3, 4, 5]],
[[ 0, 1, 2],
[ 6, 7, 8]],
[[ 3, 4, 5],
[ 6, 7, 8]],
[[ 3, 4, 5],
[ 9, 10, 11]],
[[12, 13, 14],
[15, 16, 17]],
[[12, 13, 14],
[18, 19, 20]],
[[15, 16, 17],
[18, 19, 20]],
[[15, 16, 17],
[21, 22, 23]],
[[24, 25, 26],
[27, 28, 29]],
[[24, 25, 26],
[30, 31, 32]],
[[27, 28, 29],
[30, 31, 32]],
[[27, 28, 29],
[33, 34, 35]]])
I am stuck at, as to how does np.argmax(arr, axis=0) work? I know how np.argmax(axis=0) works on 2D arrays. But this 3D one has really confused me.
My Code:
arr = np.array([[[ 1, 2, 3],
[ 4, 5, 6],
[ 7, 8, 9],
[10, 11, 12]],
[[13, 14, 15],
[16, 17, 18],
[19, 20, 21],
[22, 23, 24]],
[[25, 26, 27],
[28, 29, 30],
[31, 32, 33],
[34, 35, 36]]])
Operation:
np.argmax(arr, axis = 0)
Output:
array([[2, 2, 2],
[2, 2, 2],
[2, 2, 2],
[2, 2, 2]], dtype=int64)
FYI - I do know how np.argmax(axis=0) works on 2D arrays. But this 3D one has really confused me.
You need to understand better what is axis=0 here. It can be interpreted as height level of rectangle. So your output shows different levels of that rectangle:
level 0 level 1 level 2
[ 1, 2, 3] [13, 14, 15] [16, 17, 18]
[ 4, 5, 6] [16, 17, 18] [19, 20, 21]
[ 7, 8, 9] [19, 20, 21] [22, 23, 24]
[10, 11, 12] [22, 23, 24] [25, 16, 27]
Then argmax describes indices of levels at which max values are attained. They are:
[16, 17, 18]
[19, 20, 21]
[22, 23, 24]
[25, 16, 27]
It's definitely the upmost level (number 2) for any of these cells
so argmax of every cell is assigned to 2.
i have a 1d np array "array1d" and a 3d np array "array3d", i want to sum them so the n'th value in "array1d" will be added to each of the elements of the n'th plane in array3d.
this can be done in the following loop
for i, value in enumerate(array1d):
array3d[i] += value
question is, how can this be done in a single numpy line?
example arrays:
arr1d = np.array(range(3))
>>>array([0, 1, 2])
arr3d = np.array(range(27)).reshape(3, 3, 3)
>>>array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[ 9, 10, 11],
[12, 13, 14],
[15, 16, 17]],
[[18, 19, 20],
[21, 22, 23],
[24, 25, 26]]])
wanted result:
array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[ 10, 11, 12],
[13, 14, 15],
[16, 17, 18]],
[[20, 21, 22],
[23, 24, 25],
[26, 27, 28]]])
Use Numpy's broadcasting features:
In [23]: arr1d[:, None, None] + arr3d
Out[23]:
array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[10, 11, 12],
[13, 14, 15],
[16, 17, 18]],
[[20, 21, 22],
[23, 24, 25],
[26, 27, 28]]])
This basically copies the content of arr1d across the other two dimensions (without actually copying, it just provides a view of the memory which looks like it). Instead of None, you can also use numpy.newaxis.
Alternatively, you can also use reshape:
In [32]: arr1d.reshape(3, 1, 1) + arr3d
Out[32]:
array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[10, 11, 12],
[13, 14, 15],
[16, 17, 18]],
[[20, 21, 22],
[23, 24, 25],
[26, 27, 28]]])
I'm working with 3D matrices in numpy. I'm actually passing these matrices to C using ctypes to carry out some calculation and then getting back a result. Now the thing is, my result is correct (I did the math on paper to verify), but it's just not in a form I want it to be.
Here's an example. I have a 3D array of the form:
array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[ 9, 10, 11],
[12, 13, 14],
[15, 16, 17]],
[[18, 19, 20],
[21, 22, 23],
[24, 25, 26]]])
I need to convert it to a form where the ith columns of all 2D sub-matrices form a new 2D sub-matrix, as so:
array([[[ 0, 9, 18],
[ 3, 12, 21],
[ 6, 15, 24]],
[[ 1, 10, 19],
[ 4, 13, 22],
[ 7, 16, 25]],
[[2, 11, 20],
[5, 14, 23],
[8, 17, 26]]])
I have tried using various combinations of np.rot90, np.flipud, np.fliplr, all to no avail. Any help on this would be greatly appreciated.
Thanks a lot!
Your desired output is your initial array with the order of the axes reversed. That's how NumPy generalizes transposes to arbitrary-dimensional arrays, so you can use the T attribute for this:
In [3]: x
Out[3]:
array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[ 9, 10, 11],
[12, 13, 14],
[15, 16, 17]],
[[18, 19, 20],
[21, 22, 23],
[24, 25, 26]]])
In [4]: x.T
Out[4]:
array([[[ 0, 9, 18],
[ 3, 12, 21],
[ 6, 15, 24]],
[[ 1, 10, 19],
[ 4, 13, 22],
[ 7, 16, 25]],
[[ 2, 11, 20],
[ 5, 14, 23],
[ 8, 17, 26]]])