假设我有一个类,并且想从磁盘上并行读取几个文件,并参数化类参数。什么是最正确的方法(以及如何做)?
我考虑了线程,因为它只是I / O动作。
非并行实现(1-线程)的示例:
import pandas as pd
class DataManager(object):
def __init__(self):
self.a = None
self.b = None
self.c = None
self.d = None
self.e = None
self.f = None
def load_data(self):
self.a = pd.read_csv('a.csv')
self.b = pd.read_csv('b.csv')
self.c = pd.read_csv('c.csv')
self.d = pd.read_csv('d.csv')
self.e = pd.read_csv('e.csv')
self.f = pd.read_csv('f.csv')
if __name__ == '__main__':
dm = DataManager()
dm.load_data()
# Main thread is waiting for load_data to finish.
print("finished loading data")
答案 0 :(得分:2)
在大多数情况下,I / O操作不受CPU限制,因此使用多个进程是过大的选择。使用多个线程可能会很好,但是pb.read_csv
不仅会读取文件,还会解析它可能受到CPU限制的内容。我建议您首先使用asyncio从磁盘读取文件。这是执行此操作的代码:
import asyncio
import aiofiles
async def read_file(file_name):
async with aiofiles.open(file_name, mode='rb') as f:
return await f.read()
def read_files_async(file_names: list) -> list:
loop = asyncio.get_event_loop()
return loop.run_until_complete(
asyncio.gather(*[read_file(file_name) for file_name in file_names]))
if __name__ == '__main__':
contents = read_files_async([f'files/file_{i}.csv' for i in range(10)])
print(contents)
函数read_files_async
返回文件内容(字节缓冲区)的列表,您可以将其传递给pd.read_csv
。
我认为仅优化文件读取就足够了,但是您可以与多个进程并行解析文件内容(线程和异步不会提高解析过程的性能):
import multiprocessing as mp
NUMBER_OF_CORES = 4
pool = mp.Pool(NUMBER_OF_CORES)
pool.map(pb.read_csv, contents)
您应根据自己的机器规格设置NUMBER_OF_CORES
。
答案 1 :(得分:1)
使用Python3 ThreadPoolExecutor
的可能解决方案 from concurrent.futures import ThreadPoolExecutor
import queue
import pandas as pd
def load_data_worker(data_queue, file_name):
data_queue.put(pd.read_csv(file_name))
class DataManager(object):
def __init__(self):
self.data_queue = queue.Queue()
self.data_arr = []
def load_data(self):
with ThreadPoolExecutor() as executor:
executor.submit(load_data_woker, self.data_queue, 'a.csv')
executor.submit(load_data_woker, self.data_queue, 'b.csv')
# ...
executor.submit(load_data_woker, self.data_queue, 'f.csv')
# dumping Queue of loaded data to array
self.data_arr = list(self.data_queue.queue)
if __name__ == '__main__':
dm = DataManager()
dm.load_data()
# Main thread is waiting for load_data to finish.
print("finished loading data")