如何將二維數組作為 multiprocessing.Array 傳遞給 multiprocessing.Pool？

Question

我的目標是將父數組傳遞給mp.Pool並用2 s 填充它，同時將其分發到不同的進程。 這適用於一維數組：

import numpy as np
import multiprocessing as mp
import itertools


def worker_function(i=None):
    global arr
    val = 2
    arr[i] = val
    print(arr[:])


def init_arr(arr=None):
    globals()['arr'] = arr

def main():
    arr = mp.Array('i', np.zeros(5, dtype=int), lock=False)
    mp.Pool(1, initializer=init_arr, initargs=(arr,)).starmap(worker_function, zip(range(5)))
    print(arr[:])


if __name__ == '__main__':
    main()

輸出：

[2, 0, 0, 0, 0]
[2, 2, 0, 0, 0]
[2, 2, 2, 0, 0]
[2, 2, 2, 2, 0]
[2, 2, 2, 2, 2]
[2, 2, 2, 2, 2]

但是我怎樣才能對 x 維數組做同樣的事情呢？ 向arr添加維度：

arr = mp.Array('i', np.zeros((5, 5), dtype=int), lock=False)

產生一個錯誤：

Traceback (most recent call last):
  File "C:/Users/Artur/Desktop/RL_framework/test2.py", line 23, in <module>
    main()
  File "C:/Users/Artur/Desktop/RL_framework/test2.py", line 17, in main
    arr = mp.Array('i', np.zeros((5, 5), dtype=int), lock=False)
  File "C:\Users\Artur\anaconda3\envs\RL_framework\lib\multiprocessing\context.py", line 141, in Array
    ctx=self.get_context())
  File "C:\Users\Artur\anaconda3\envs\RL_framework\lib\multiprocessing\sharedctypes.py", line 88, in Array
    obj = RawArray(typecode_or_type, size_or_initializer)
  File "C:\Users\Artur\anaconda3\envs\RL_framework\lib\multiprocessing\sharedctypes.py", line 67, in RawArray
    result.__init__(*size_or_initializer)
TypeError: only size-1 arrays can be converted to Python scalars

更改dtype的arr也沒有幫助。

Answer 1

您不能直接將multiprocessing.Array用作二維數組，但在一維內存中，無論如何，第二維只是一種錯覺:)。

幸運的是 numpy 允許從緩沖區讀取數組並對其進行整形，而無需復制它。 在下面的演示中，我只使用了一個單獨的鎖，這樣我們就可以逐步觀察所做的更改，目前沒有競爭條件。

import multiprocessing as mp
import numpy as np    

def worker_function(i):
    global arr, arr_lock
    val = 2
    with arr_lock:
        arr[i, :i+1] = val
        print(f"{mp.current_process().name}\n{arr[:]}")


def init_arr(arr, arr_lock=None):
    globals()['arr'] = np.frombuffer(arr, dtype='int32').reshape(5, 5)
    globals()['arr_lock'] = arr_lock


def main():
    arr = mp.Array('i', np.zeros(5 * 5, dtype='int32'), lock=False)
    arr_lock = mp.Lock()

    mp.Pool(2, initializer=init_arr, initargs=(arr, arr_lock)).map(
        worker_function, range(5)
    )

    arr = np.frombuffer(arr, dtype='int32').reshape(5, 5)
    print(f"{mp.current_process().name}\n{arr}")


if __name__ == '__main__':
    main()

輸出：

ForkPoolWorker-1
[[2 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]]
ForkPoolWorker-2
[[2 0 0 0 0]
 [2 2 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]]
ForkPoolWorker-1
[[2 0 0 0 0]
 [2 2 0 0 0]
 [2 2 2 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]]
ForkPoolWorker-2
[[2 0 0 0 0]
 [2 2 0 0 0]
 [2 2 2 0 0]
 [2 2 2 2 0]
 [0 0 0 0 0]]
ForkPoolWorker-1
[[2 0 0 0 0]
 [2 2 0 0 0]
 [2 2 2 0 0]
 [2 2 2 2 0]
 [2 2 2 2 2]]
MainProcess
[[2 0 0 0 0]
 [2 2 0 0 0]
 [2 2 2 0 0]
 [2 2 2 2 0]
 [2 2 2 2 2]]

Process finished with exit code 0