python 多进程并发demo

outline

下午需要简单处理一份数据，就直接随手写脚本处理了，但发现效率太低，速度太慢，就改成多进程了；

程序涉及计算、文件读写，鉴于计算内容挺多的，就用多进程了（计算密集）。

代码

import pandas as pd
from pathlib import Path
from concurrent.futures import ProcessPoolExecutor

parse_path = '/data1/v-gazh/CRSP/dsf_full_fields/parse'
source_path = '/data1/v-gazh/CRSP/dsf_full_fields/2th_split'  # 目录中有3.3W个csv文件，串行的话，效率大打折扣


def parseData():
    source_path_list = list(Path(source_path).glob('*.csv'))
    multi_process = ProcessPoolExecutor(max_workers=20)
    multi_results = multi_process.map(func, source_path_list)


def func(p):
    source_p = str(p)
    parse_p = str(p).replace('2th_split', 'parse')
    df = pd.read_csv(source_p)
    df['date'] = pd.to_datetime(df['date'].astype(str)).dt.date
    df.sort_values(['date'], inplace=True)
    # 处理close为负的值(abs)，添加status标识
    df['is_close'] = df['PRC'].map(lambda x: 0 if x < 0 or pd.isna(x) else 1)
    df['PRC'] = df['PRC'].abs()
    df.rename(columns={'CFACPR': 'factor'}, inplace=True)
    df['adj_low'] = df['BIDLO'] * df['factor']
    df['adj_high'] = df['ASKHI'] * df['factor']
    df['adj_close'] = df['PRC'] * df['factor']
    df['adj_open'] = df['OPENPRC'] * df['factor']
    df['adj_volume'] = df['VOL'] / df['factor']
    # calc change
    df['change'] = df['adj_close'].diff(1) / df['adj_close'].shift(1)
　　 df.drop_duplicates(inplace=True)
    df.to_csv(parse_p, index=False)


parseData()