I am wondering the possibility of turning a for loop with several if-else statements into vectorize version in Python. Suppose I have several numpy arrays a,b,c,d , where a,b,c,d have the same length. I need to write the following for loop to calculate the value of c and ‘d’

    for i in range(2, len(a)):
        if (
            a[i - 2] < b[i - 2]
            and a[i - 1] > b[i - 1]
            and c[i - 1] == 0
        ):
            d[i] = -1
            c[i] = c[i - 1] - 1
        elif (
            a[i - 2] > b[i - 2]
            and a[i - 1] < b[i - 1]
            and c[i - 1] == 0
        ):
            d[i] = 1
            c[i] = c[i - 1] + 1
        else: 
            d[i] = 0  
            c[i] = c[i - 1]

The catch is that the size of arrays is 10^5 and I have to execute this for-loop for 10^3' times, the time complexity is O(10^8)` which is very slow and computational heavy to use for loop with if-else statement. I am wondering is there a vectorized or faster way to do these operation in Python?

I have ask ChatGPT for helps, it suggests that turning the if , elif conditions to

condition1 = (
    (a[:-2] < b[:-2]) &
    (a[1:-1] > b[1:-1]) &
    (d 1:-1] == 0)
)

condition2 = (
    (a[:-2] > b[:-2]) &
    (a[1:-1] < b[1:-1]) &
    (d 1:-1] == 0)
)

, respectively. Then, it uses np.where to execute the conditions

c[2:] = np.where(condition1, -1,0)
d[2:] = np.where(condition1, d[1:-1] - 1, d[1:-1])

c[2:] = np.where(condition2, 1, 0)
d[2:] = np.where(condition2, d1:-1] + 1, d[1:-1])

however, these results are not correct comparing the for-loop result. Beside, I don’t know where is the condition for else statement.