我有一些存储在numpy数组中的参数集,我将它们放入了一个多进程队列中,但是当工作进程接收到它们时它们变得混乱。下面是我的代码,以说明我的问题和疑问。
import numpy as np
from multiprocessing import Process, Queue
NUMBER_OF_PROCESSES = 2
def worker(input, output):
for args in iter(input.get, 'STOP'):
print('Worker receives: ' + repr(args))
id, par = args
# simulate a complex task, and return result
result = par['A'] * par['B']
output.put((id, result))
# Define parameters to process
parameters = np.array([
(1.0, 2.0),
(3.0, 3.0)], dtype=[('A', 'd'), ('B', 'd')])
# Create queues
task_queue = Queue()
done_queue = Queue()
# Submit tasks
for id, par in enumerate(parameters):
obj = ('id_' + str(id), par)
print('Submitting task: ' + repr(obj))
task_queue.put(obj)
# Start worker processes
for i in range(NUMBER_OF_PROCESSES):
Process(target=worker, args=(task_queue, done_queue)).start()
# Get unordered results
results = {}
for i in range(len(parameters)):
id, result = done_queue.get()
results[id] = result
# Tell child processes to stop
for i in range(NUMBER_OF_PROCESSES):
task_queue.put('STOP')
print('results: ' + str(results))
在一台64位CentOS电脑上,使用numpy 1.4.1和Python 2.6.6,我的输出结果如下:
Submitting task: ('id_0', (1.0, 2.0))
Submitting task: ('id_1', (3.0, 3.0))
Worker receives: ('id_0', (2.07827093387802e-316, 6.9204740511333381e-310))
Worker receives: ('id_1', (0.0, 1.8834810076011668e-316))
results: {'id_0': 0.0, 'id_1': 0.0}
正如所示,当提交任务时,具有numpy记录数组的元组处于良好状态,但是当worker接收参数时,它们会乱码,并且结果不正确。我在
multiprocessing
documentation中读到,“代理方法的参数是可picklable的”。据我所知,numpy数组是完全可picklable的:>>> import pickle
>>> for par in parameters:
... print(pickle.loads(pickle.dumps(par)))
...
(1.0, 2.0)
(3.0, 3.0)
我的问题是为什么参数在worker中没有被正确接收?我该如何将numpy记录数组的一行传递给worker?