使用适当的步长和步长范围创建numpy数组

Question

Hi all I need to create specific array contains random integers from certain range with step.

Output need to by like this:

[[0 3 2 1]
 [1 2 0 3]
 [2 0 3 1]
    ...
 [10 7 8 9]]

First digit in each row need to be a certain value [0,1,2,...] next 3 digits in the row need to be random from range (0,3) for first row (without 0), (0,3) for the second (without 1), (6,10) for last one (without 10). Generally (x, y) without fixed digit where x = fixed number - 4 but not less than the fixed number, y = fixed number + 4 but not more than the fixed number .

Range of random numbers need to be 4 digits without repeats in certain direction.

Another visualization: fixed numbers = [0,1,2,3,..] random range = (0,3) + offset

            array              range - without

[[0, random numbers in range]  (0,3) - 0
 [1, random numbers in range]  (0,3) - 1
 [2, random numbers in range]  (0,3) - 2
       ...
 [6, random numbers in range]  (2,6) - 6
 [7, random numbers in range]  (3,7) - 7
       ...
 [n, random numbers in range]]  (n-4,n) - n

After some manipulations i wrote below code, which is pretty fast but i think that it can by done much better and faster.

def winning_matrix():
    ######## create beginning 4 vector array ########
    # create vector with counts to merge
    add_vector_beginning = np.arange(4)[:, None]
    # create zeros array for data input
    add_array = np.zeros(shape=(4, 3))
    # fill up add array
    add_array[0] = np.array([1, 2, 3])
    add_array[1] = np.array([0, 2, 3])
    add_array[2] = np.array([0, 1, 3])
    add_array[3] = np.array([0, 1, 2])
    # convert array values to integers
    add_array = add_array.astype(int)

    # mix array values
    mix_array = np.zeros(shape=(4, 3))
    for i in range(4):
         mix_array[i] = np.random.choice(add_array[i], replace=False, size=3)
    # convert array values to integers
    mix_array_int = mix_array.astype(int)
    # merge array with proper vector
    add_matrix = np.column_stack((add_vector_beginning, mix_array_int))
    #### create main winning array #######
    # create add vector with proper count values
    add_vector = np.arange(vectors_number - 4)[:, None] + 4
    # create offset for main array
    offset = np.arange(vectors_number - 4)[:, None] + 1
    # create random array
    winning_matrix_random = np.random.rand(vectors_number - 4, 3)
    # convert random array values in array into positions intigers
    winning_convert = winning_matrix_random.argsort(1)
    # add offset to array
    winning_matrix_raw = winning_convert + offset
    # merge add vector with winning array
    winning_matrix = np.column_stack((add_vector, winning_matrix_raw))

    #### merge two created arrays into one #########
    for i in range(4):
         winning_matrix = np.insert(winning_matrix, i, add_matrix[i], axis=0)
    return winning_matrix

Any suggestions how to speed up?

Answer 1

Looks like the choices for each line are independent, depending on a n parameter (or two). If so I'd focus on writing a function that generates one row given n .

For that row, generate a range, eg x = np.arange(m, m+4) , remove n , then do a np.random.choise(x, size=3, replace=False) . I haven't read your description with enough care to get all the details.

Then just run this function for each row, accumulating the results in a list.

def arow(n,m,k=4):
    x = list(range(m,m+k))
    i = x.pop(x.index(n))
    y = np.random.choice(x, replace=False, size=3)
    y = [i]+y.tolist()
    return y

In [34]: arow(0,0,4)
Out[34]: [0, 1, 2, 3]
In [35]: arow(1,0,4)
Out[35]: [1, 2, 3, 0]
In [39]: arow(10,7,4)
Out[39]: [10, 8, 7, 9]

In [47]: k=4
    ...: alist = []
    ...: for n in range(11):
    ...:     m = max(0,n-k+1)
    ...:     r = arow(n,m,k)
    ...:     alist.append(r)
    ...: np.array(alist)

Out[47]: 
array([[ 0,  2,  3,  1],
       [ 1,  0,  3,  2],
       [ 2,  3,  1,  0],
       [ 3,  1,  2,  0],
       [ 4,  3,  2,  1],
       [ 5,  4,  3,  2],
       [ 6,  3,  4,  5],
       [ 7,  6,  4,  5],
       [ 8,  7,  6,  5],
       [ 9,  8,  6,  7],
       [10,  8,  9,  7]])