Python 多进程计算示例
import logging
from multiprocessing import Pool
import time
import pandas as pd
# 计算注册买单时间 (时间较长,推荐分割数据后用多进程进行计算)
bins = [0,1,15,30,45,60,75,90,105,120,135,150,165,180,360]
start = time.time()
def df_mp(df, row_start, row_end):
"""多进程函数"""
# 分拆数据
df_sub = df[row_start : row_end]
df_sub = df_sub[df_sub['JoinTime'] < '2021-01-01']
# 计算注册买单时间
df_sub['tranTime'] = df_sub.apply(lambda x: x['FirstPaymentTime'] - x['JoinTime'], axis=1)
del df_sub['FirstPaymentTime']
df_sub['tranTime'] = df_sub['tranTime'].apply(lambda x: x.days)
df_sub.fillna(181.0, inplace=True)
df_sub['tranTime'] = df_sub['tranTime'] + 1
df_sub['tranTime_bin'] = pd.cut(df_sub['tranTime'], bins)
return df_sub
if __name__ == '__main__':
pool = Pool()
batch_uids = 500000 # 每个数据自己的数量
batchs = int(len(allData2) / batch_uids) + 1
res_l = []
for i in range(batchs):
m = i * batch_uids # 切片始
n = (i + 1) * batch_uids # 切片终
res = pool.apply_async(df_mp, args=(allData2, m, n,)) # 此处不能用get方法,会阻塞进程池
res_l.append(res)
print("==============================>")
pool.close()
pool.join() #调用join之前,先调用close函数,否则会出错。执行完close后不会有新的进程加入到pool,join函数等待所有子进程>结束
df = res_l[0].get()
for res in res_l[1:]:
df = df.append(res.get())
logger.info("all done. cost: %3f 秒" % (time.time() - start))