我有兩個(gè)數(shù)據(jù)框left = pd.DataFrame([['A', 10, datetime(2020, 5, 17, 20, 12, 28)],\ ['B', 15, datetime(2020, 5, 17, 16, 22, 45)],\ ['C', 20, datetime(2020, 5, 17, 12, 45, 12)],\ ['D', 25, datetime(2020, 5, 17, 13, 57, 44)]],\ columns = ['Letter_l', 'Int_l', 'Datetime_l'])和right = pd.DataFrame([['A', 20, datetime(2020, 5, 17, 20, 12, 35)],\ ['B', 30, datetime(2020, 5, 17, 18, 45, 25)],\ ['C', 40, datetime(2020, 5, 17, 12, 45, 20)],\ ['D', 50, datetime(2020, 5, 17, 18, 16, 44)]],\ columns = ['Letter_r', 'Int_r', 'Datetime_r'])我想加入三列中的兩列：Letter和Datetime。對(duì)于日期時(shí)間，我不是在尋找精確匹配，而是在尋找正確數(shù)據(jù)幀的最大差異 10 秒。我知道如何在正常條件下加入：merged_df = pd.merge(left=left, right=right, how='left',\ left_on=['Letter_l'], right_on=['Letter_r'])但我正在尋找的輸出是：|-----|--------------|-----------|--------------------|--------------|-----------|--------------------|| | Letter_l | Int_l | datetime_l | Letter_r | Int_r | datetime_r | |-----|--------------|-----------|--------------------|--------------|-----------|--------------------|| 0 | A | 10 | 2020-05-17 20:12:28| A | 20 | 2020-05-17 20:12:35|| 1 | C | 20 | 2020-05-17 12:45:12| C | 40 | 2020-05-17 12:45:20||-----|--------------|-----------|--------------------|--------------|-----------|--------------------|這可以使用標(biāo)準(zhǔn)來完成嗎pd.merge？當(dāng)然我可以嘗試使用sqlite3
1 回答
慕神8447489
TA貢獻(xiàn)1780條經(jīng)驗(yàn) 獲得超1個(gè)贊
使用on執(zhí)行asof合并，容差為秒：pd.merge_asofDateTime10
df = pd.merge_asof(left.sort_values('Datetime_l'),
? ? ? ? ? ? ? ? ? ?right.sort_values('Datetime_r'),
? ? ? ? ? ? ? ? ? ?left_by='Letter_l', right_by='Letter_r',
? ? ? ? ? ? ? ? ? ?left_on='Datetime_l', right_on='Datetime_r',
? ? ? ? ? ? ? ? ? ?direction='nearest', tolerance=pd.Timedelta(seconds=10))\
? ? ? ?.dropna(subset=['Letter_r'])
結(jié)果：
? Letter_l? Int_l? ? ? ? ? Datetime_l Letter_r? Int_r? ? ? ? ? Datetime_r
0? ? ? ? C? ? ?20 2020-05-17 12:45:12? ? ? ? C? ?40.0 2020-05-17 12:45:20
3? ? ? ? A? ? ?10 2020-05-17 20:12:28? ? ? ? A? ?20.0 2020-05-17 20:12:35
添加回答
舉報(bào)
0/150
提交
取消