更改最接近Numpy数组中的值的n个元素的符号

Question

I wish to change the sign of the n elements in a numpy array that lies closest to a certain value, but not less. That is, the elements must be equal or greater than the value. Is there any fast Numpy methods that can do this efficiently with large arrays?

The code that I have now takes n values that are higher or equal, but not closest, which is "okay", but not ideal for my results.

def update(arr, n, value):
    updated = 0
    i = 0
    while updated < n:
        if arr[i] >= value: # just a random value above "value"
            arr[i] = -arr[i]
            updated +=1
        i += 1

arr = np.array([9, 8, 2, -4, 3, 4])
n = 3
value = 2
update(arr, n, value)

gives me

arr = np.array([-9, -8, -2, -4, 3, 4])

when I instead want

arr = np.array([9, 8, -2, -4, -3, -4])

Answer 1

I'm not updating the array in place, but I'd do something like:

def update(arr, n, value):
    arr_copy = arr.copy()
    diffs = arr - value
    absolute_diffs = np.abs(diffs)
    update_indeces = np.argpartition(absolute_diffs, n)[:n]
    arr_copy[update_indeces] *= -1
    return arr_copy

Answer 2

You can use argpartition :

arr = np.random.random(20)
value = 0.5
n = 4

nl = np.count_nonzero(arr<value)
closest = np.argpartition(arr, (nl, nl+n-1))[nl:nl+n]
arr[closest] = -arr[closest]
arr
# array([ 0.33697627,  0.42607914, -0.63703314, -0.57517234,  0.82674228,
#        -0.52929285,  0.64776714,  0.25609886,  0.24681445,  0.2486823 ,
#         0.76740245,  0.02368603,  0.21498096, -0.51033841,  0.19901665,
#         0.30939207,  0.69036139,  0.83178506,  0.97243443,  0.47620492])

Answer 3

This should work:

def flip_some(a, n, value):
    more_than = (a >= value)
    first_n_elements = (a < np.sort(a[more_than])[n])
    return np.where(more_than & first_n_elements, -a, a)

print(flip_some(np.array([9, 8, 2, -4, 3, 4]), 3, 2))
print(flip_some(np.arange(10), 2, 5))

Output:

[ 9 -8  -2 -4 -3 -4]
[ 0  1  2  3  4 -5 -6 -7  8  9]