We Keep Coding

Numpy get secondary diagonal with offset=1 and change the values

I have this 6x6 matrix filled with 0s. I got the secondary diagonal in sec_diag. The thing I am trying to do is to change the values of above the sec_diag inside the matrix with the odds numbers from 9-1 [9,7,5,3,1]
import numpy as np
x = np.zeros((6,6), int)
sec_diag = np.diagonal(np.fliplr(x), offset=1)
The result should look like this:
[[0,0,0,0,9,0],
[0,0,0,7,0,0],
[0,0,5,0,0,0],
[0,3,0,0,0,0],
[1,0,0,0,0,0],
[0,0,0,0,0,0]]
EDIT: np.fill_diagonal isn't going to work.

You should use roll
x = np.zeros((6,6),dtype=np.int32)
np.fill_diagonal(np.fliplr(x), [9,7,5,3,1,0])
xr = np.roll(x,-1,axis=1)
print(xr)
Output
[[0 0 0 0 9 0]
[0 0 0 7 0 0]
[0 0 5 0 0 0]
[0 3 0 0 0 0]
[1 0 0 0 0 0]
[0 0 0 0 0 0]]

Maybe you should try with a double loop

How to resize a Numpy array to add/replace rows with combinations determined by the values in each row of the array

So I have a program that at some point creates random arrays and I have perform an operation which is to add rows while replacing other rows based on the values found in the rows. One of the random arrays will look something like this but keep in mind that it could randomly vary in size ranging from 3x3 up to 10x10:
0 2 0 1
1 0 0 1
1 0 2 1
2 0 1 2
For every row that has at least one value equal to 2 I need to remove/replace the row and add some more rows. The number of rows added will depend on the number of combinations possible of 0s and 1s where the number of digits is equal to the number of 2s counted in each row. Each added row will introduce one of these combinations in the positions where the 2s are located. The result that I'm looking for will look like this:
0 1 0 1 # First combination to replace 0 2 0 1
0 0 0 1 # Second combination to replace 0 2 0 1 (Only 2 combinations, only one 2)
1 0 0 1 # Stays the same
1 0 1 1 # First combination to replace 1 0 2 1
1 0 0 1 # Second combination to replace 1 0 2 1 (Only 2 combinations, only one 2)
0 0 1 0 # First combination to replace 2 0 1 2
0 0 1 1 # Second combination to replace 2 0 1 2
1 0 1 1 # Third combination to replace 2 0 1 2
1 0 1 0 # Fourth combination to replace 2 0 1 2 (4 combinations, there are two 2s)
If you know a Numpy way of accomplishing this I will be grateful.

You can try the following. Create a sample array:
import numpy as np
np.random.seed(5)
a = np.random.randint(0, 3, (4, 4))
print(a)
This gives:
[[2 1 2 2]
[0 1 0 0]
[2 0 2 0]
[0 1 1 0]]
Compute the output array:
ts = (a == 2).sum(axis=1)
r = np.hstack([np.array(np.meshgrid(*[[0, 1]] * t)).reshape(t, -1).T.ravel() for t in ts if t])
out = np.repeat(a, 2**ts, axis=0)
out[out == 2] = r
print(out)
Result:
[[0 1 0 0]
[0 1 0 1]
[1 1 0 0]
[1 1 0 1]
[0 1 1 0]
[0 1 1 1]
[1 1 1 0]
[1 1 1 1]
[0 1 0 0]
[0 0 0 0]
[1 0 0 0]
[0 0 1 0]
[1 0 1 0]
[0 1 1 0]]

Not the prettiest code but it does the job. You could clean up the itertools calls but this lets you see how it works.
import numpy as np
import itertools
X = np.array([[0, 2, 0, 1],
[1, 0, 0, 1],
[1, 0, 2, 1],
[2, 0, 1, 2]])
def add(X_,Y):
if Y.size == 0:
Y = X_
else:
Y = np.vstack((Y, X_))
return(Y)
Y = np.array([])
for i in range(len(X)):
if 2 not in X[i,:]:
Y = add(X[i,:], Y)
else:
a = np.where(X[i,:]==2)[0]
n = [[i for i in itertools.chain([1, 0])] for _ in range(len(a))]
m = list(itertools.product(*n))
for j in range(len(m)):
M = 1 * X[i,:]
u = list(m[j])
for k in range(len(a)):
M[a[k]] = u[k]
Y = add(M, Y)
print(Y)
#[[0 1 0 1]
# [0 0 0 1]
# [1 0 0 1]
# [1 0 1 1]
# [1 0 0 1]
# [1 0 1 1]
# [1 0 1 0]
# [0 0 1 1]
# [0 0 1 0]]

Write functions resilient to variable dimension array

I'm struggling when writing a function that would seemlessly apply to any numpy arrays whatever its dimension.
At one point in my code, I have boolean arrays that I consider as mask for other arrays (0 = not passing, 1 = passing).
I would like to "enlarge" those mask arrays by overriding zeros adjacent to ones on a defined range.
Example :
input = [0,0,0,0,0,1,0,0,0,0,1,0,0,0]
enlarged_by_1 = [0,0,0,0,1,1,1,0,0,1,1,1,0,0]
enlarged_by_2 = [0,0,0,1,1,1,1,1,1,1,1,1,1,0]
input = [[0,0,0,1,0,0,1,0],
[0,1,0,0,0,0,0,0],
[0,0,0,0,0,0,1,0]]
enlarged_by_1 = [[0,0,1,1,1,1,1,1],
[1,1,1,0,0,0,0,0],
[0,0,0,0,0,1,1,1]]
This is pretty straighforward when inputs are 1D.
However, I would like this function to take seemlessy 1D, matrix, 3D, and so on.
So for a matrix, the same logic would be applied to each lines.
I read about ellipsis, but it does not seem to be applicable in my case.
Flattening the input applying the logic and reshaping the array would lead to possible contamination between individual arrays.
I do not want to go through testing the shape of input numpy array / recursive function as it does not seems very clean to me.
Would you have some suggestions ?

The operation that you are described seems very much like a convolution operation followed by clipping to ensure that values remain 0 or 1.
For your example input:
import numpy as np
input = np.array([0,0,0,0,0,1,0,0,0,0,1,0,0,0], dtype=int)
print(input)
def enlarge_ones(x, k):
mask = np.ones(2*k+1, dtype=int)
return np.clip(np.convolve(x, mask, mode='same'), 0, 1).astype(int)
print(enlarge_ones(input, k=1))
print(enlarge_ones(input, k=3))
which yields
[0 0 0 0 0 1 0 0 0 0 1 0 0 0]
[0 0 0 0 1 1 1 0 0 1 1 1 0 0]
[0 0 1 1 1 1 1 1 1 1 1 1 1 1]
numpy.convolve only works for 1-d arrays. However, one can imagine a for loop over the number of array dimensions and another for loop over each array. In other words, for a 2-d matrix first operate on every row and then on every column. You get the idea for nd-array with more dimensions. In other words the enlarge_ones would become something like:
def enlarge_ones(x, k):
n = len(x.shape)
if n == 1:
mask = np.ones(2*k+1, dtype=int)
return np.clip(np.convolve(x, mask, mode='same')[:len(x)], 0, 1).astype(int)
else:
x = x.copy()
for d in range(n):
for i in np.ndindex(x.shape[:-1]):
x[i] = enlarge_ones(x[i], k) # x[i] is 1-d
x = x.transpose(list(range(1, n)) + [0])
return x
Note the use of np.transpose to rotate the dimensions so that np.convolve is applied to the 1-d along each dimension. This is exactly n times, which returns the matrix to original shape at the end.
x = np.zeros((3, 5, 7), dtype=int)
x[1, 2, 2] = 1
print(x)
print(enlarge_ones(x, k=1))
[[[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]]
[[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 1 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]]
[[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]
[0 0 0 0 0 0 0]]]
[[[0 0 0 0 0 0 0]
[0 1 1 1 0 0 0]
[0 1 1 1 0 0 0]
[0 1 1 1 0 0 0]
[0 0 0 0 0 0 0]]
[[0 0 0 0 0 0 0]
[0 1 1 1 0 0 0]
[0 1 1 1 0 0 0]
[0 1 1 1 0 0 0]
[0 0 0 0 0 0 0]]
[[0 0 0 0 0 0 0]
[0 1 1 1 0 0 0]
[0 1 1 1 0 0 0]
[0 1 1 1 0 0 0]
[0 0 0 0 0 0 0]]]

using 1's in the arrays to make a shape like '+' using numpy in python

Numpy Three Four Five Dimensional Array in Python
Input 1: 3
Output 1:
[[0 1 0]
[1 1 1]
[0 1 0]]
Input 2:5
Output 1:
[[0 0 1 0 0]
[0 0 1 0 0]
[1 1 1 1 1]
[0 0 1 0 0]
[0 0 1 0 0]]
Notice that the 1s in the arrays make a shape like +.
My logic is shown below
a=np.zeros((n,n),dtype='int')
a[-3,:] = 1
a[:,-3] = 1 print(a)
This logic is only working for five dimensional array but not for three dimensional array.
can someone assist me to get the expected output for both three and five dimensional array using np.zeros & integer division //

As you can see, n//2 = 3 when n=5. So, that's the solution to your question as see here:
import numpy as np
def create_plus_matrix(n):
a = np.zeros((n,n),dtype='int')
a[-n//2,:] = 1
a[:,-n//2] = 1
return a
So, let's try it out:
>>> create_plus_matrix(3)
[[0 1 0]
[1 1 1]
[0 1 0]]
>> create_plus_matrix(5)
[[0 0 1 0 0]
[0 0 1 0 0]
[1 1 1 1 1]
[0 0 1 0 0]
[0 0 1 0 0]]

Do this
import numpy as np
def plus(size):
a = np.zeros([size,size], dtype = int)
a[int(size/2)] = np.ones(size)
for i in a:
i[int(size/2)] = 1
return a
print(plus(3)) //3 is the size
//Output
[[0 1 0]
[1 1 1]
[0 1 0]]

How to pad an array non-symmetrically (e.g., only from one side)?

There is an example in Numpy's documentation for padding 2D arrays with constants:
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
but it works for symmetric paddings only. For instance, np.pad(a, ((2, 2), (1, 1)), pad_with, padder=0) gives:
[[0 0 0 0 0]
[0 0 0 0 0]
[0 1 1 1 0]
[0 1 1 1 0]
[0 0 0 0 0]
[0 0 0 0 0]]
Question: How can I pad the array only from specific sides (i.e., only left and top sides)? Like this:
[[0 0 0 0]
[0 0 0 0]
[0 1 1 1]
[0 1 1 1]]

It turns our that with a simple change we can achieve that:
def pad_with(vector, pad_width, iaxis, kwargs):
pad_value = kwargs.get('padder', 0)
vector[:pad_width[0]] = pad_value
if pad_width[1] != 0: # <-- the only change (0 indicates no padding)
vector[-pad_width[1]:] = pad_value
Here are some examples:
Padding 1 row of zeros (only) to the top:
>>> np.pad(a, ((1, 0), (0, 0)), pad_with, padder=0)
[[0 0 0]
[1 1 1]
[1 1 1]]
Padding 2 rows of zeros, both to the left and right:
np.pad(a, ((0, 0), (2, 2)), pad_with, padder=0)
[[0 0 1 1 1 0 0]
[0 0 1 1 1 0 0]]
and so on.