[FIXED] python Numpy. Find the accumulative sum of an array

Issue

I am hoping to achieve the following and return 'result'. Is there a way to make the following code more efficient? Thank you in advance. The problem for me is that x.shape[1] would not just be 2 anymore in my real problem. It can be 100 or even more. Hence accessing idx would be much more difficult. I am not sure how I should improve this.

import numpy as np

np.random.seed(0)
x = np.random.randint(0, high = 50, size = (5, 2))
y = np.random.randint(0, high = 50, size = (5))

result = np.zeros(5)
for i in range(len(x)):
    idx =( x[:,0]<= x[i,0]) & (x[:,1]<= x[i,1])
    #print(y[idx])
    result[i] = np.sum(y[idx])

Solution

Use broadcasting:

# compare x to itself and aggregate with all
idx = (x <= x[:, None]).all(axis=2)

# broadcast y using the boolean mask
# to have the value if True and 0 if False
# then sum
result = (y * idx).sum(axis=0)

Output: array([23, 30, 48, 72, 24])

Used input (with np.random.seed(0)):

# x
array([[44, 47],
       [ 0,  3],
       [ 3, 39],
       [ 9, 19],
       [21, 36]])

# y
array([23,  6, 24, 24, 12])

Intermediates:

# idx
array([[ True,  True,  True,  True,  True],
       [False,  True, False, False, False],
       [False,  True,  True, False, False],
       [False,  True, False,  True, False],
       [False,  True, False,  True,  True]])

# y * idx
array([[23,  6, 24, 24, 12],
       [ 0,  6,  0,  0,  0],
       [ 0,  6, 24,  0,  0],
       [ 0,  6,  0, 24,  0],
       [ 0,  6,  0, 24, 12]])

operations other than sum

If you want to compute something other than a sum you might be annoyed by the 0s and would prefer NaNs.

In this case better use where:

np.where(idx, y, np.nan)
# array([[23.,  6., 24., 24., 12.],
#        [nan,  6., nan, nan, nan],
#        [nan,  6., 24., nan, nan],
#        [nan,  6., nan, 24., nan],
#        [nan,  6., nan, 24., 12.]])

np.nansum(np.where(idx, y, np.nan), axis=0)
# array([23., 30., 48., 72., 24.])

optimization of the loop approach

While broadcasting is better, you could also have improved the slicing part in your loop to act on all columns at once:

for i in range(len(x)):
    idx = (x<=x[i]).all(1)
    result[i] = np.sum(y[idx])

Answered By - mozway