Question

我正在尝试一次处理多个文件，其中每个文件将生成数据块，以同时馈入一定大小限制的队列。例如，如果有5个文件，每个文件包含一百万个元素，我想将每个文件中的100个元素提供给另一个生成器，该生成器一次生成500个元素。

这是到目前为止我一直在尝试的方法，但是遇到can't pickle generator错误：

import os
from itertools import islice
import multiprocessing as mp
import numpy as np

class File(object):
    def __init__(self, data_params):
        data_len = 100000
        self.large_data = np.array([data_params + str(i) for i in np.arange(0, data_len)])
    def __iter__(self):
        for i in self.large_data:
            yield i

def parse_file(file_path):
    # differnt filepaths yeild different data obviously
    # here we just emulate with something silly
    if file_path == 'elephant_file':
        p = File(data_params = 'elephant')
    if file_path == 'number_file':
        p = File(data_params = 'number')
    if file_path == 'horse_file':
        p = File(data_params = 'horse')


    yield from p

def parse_dir(user_given_dir, chunksize = 10):
    pool = mp.Pool(4)
    paths = ['elephant_file', 'number_file', 'horse_file'] #[os.path.join(user_given_dir, p) for p in os.listdir(user_given_dir)]

    # Works, but not simultaneously on all paths
#     for path in paths:
#         data_gen = parse_file(path)
#         parsed_data_batch = True
#         while parsed_data_batch:
#             parsed_data_batch = list(islice(data_gen, chunksize))
#             yield parsed_data_batch

    # Doesn't work
    for objs in pool.imap(parse_file, paths, chunksize = chunksize):
        for o in objs:
            yield o

it = parse_dir('.')
for ix, o in enumerate(it):
    print(o) # hopefully just prints 10 elephants, horses and numbers
    if ix>2: break

有人知道如何获得所需的行为吗？

Answer 1

对于泡菜错误：

parse_file是生成器，而不是常规函数，因为它在内部使用了yield。

multiprocessing需要一个函数作为任务来执行。因此，您应该在yield from p

return p

parse_file()

如果要从所有文件中逐块读取记录，请尝试在zip中使用parse_dir()：

iterators = [
    iter(e) for e in pool.imap(parse_file, paths, chunksize=chunksize)
]

while True:
    batch = [
        o for i in iterators
        for _, o in zip(range(100), i)  # e.g., 100
    ]
   if batch:
        yield batch
    else:
        return

在多个文件上使用生成器进行多处理，并围绕TypeError（“无法使生成器对象腌制”）进行处理

1 个答案: