[英]Most efficient way to use resample on groupby with start and end datetime while preserving certain columns - and calculate statistics after that
[英]Efficient way of filtering by datetime in groupby
鑒於由DataFrame
生成的DataFrame
:
import numpy as np
import pandas as pd
from datetime import timedelta
np.random.seed(0)
rng = pd.date_range('2015-02-24', periods=14, freq='9H')
ids = [1]*5 + [2]*2 + [3]*7
df = pd.DataFrame({'id': ids, 'time_entered': rng, 'val': np.random.randn(len(rng))})
df
:
id time_entered val
0 1 2015-02-24 00:00:00 1.764052
1 1 2015-02-24 09:00:00 0.400157
2 1 2015-02-24 18:00:00 0.978738
3 1 2015-02-25 03:00:00 2.240893
4 1 2015-02-25 12:00:00 1.867558
5 2 2015-02-25 21:00:00 -0.977278
6 2 2015-02-26 06:00:00 0.950088
7 3 2015-02-26 15:00:00 -0.151357
8 3 2015-02-27 00:00:00 -0.103219
9 3 2015-02-27 09:00:00 0.410599
10 3 2015-02-27 18:00:00 0.144044
11 3 2015-02-28 03:00:00 1.454274
12 3 2015-02-28 12:00:00 0.761038
13 3 2015-02-28 21:00:00 0.121675
對於每個id
,我需要從最新的time_entered
刪除超過 24 小時(1 天)的time_entered
,對於該id
。 我目前的解決方案:
def custom_transform(x):
datetime_from = x["time_entered"].max() - timedelta(days=1)
x = x[x["time_entered"] > datetime_from]
return x
df.groupby("id").apply(lambda x: custom_transform(x)).reset_index(drop=True)
它給出了正確的、預期的輸出:
id time_entered val
0 1 2015-02-24 18:00:00 0.978738
1 1 2015-02-25 03:00:00 2.240893
2 1 2015-02-25 12:00:00 1.867558
3 2 2015-02-25 21:00:00 -0.977278
4 2 2015-02-26 06:00:00 0.950088
5 3 2015-02-28 03:00:00 1.454274
6 3 2015-02-28 12:00:00 0.761038
7 3 2015-02-28 21:00:00 0.121675
但是,我的真實數據是幾千萬行,還有幾十萬個唯一ID,因此,這個解決方案是不可行的(需要很長時間)。
有沒有更有效的方法來過濾數據? 我欣賞所有的想法!
一般來說,避免groupby().apply()
因為它沒有跨組向量化,更不用說如果你像你的情況一樣返回新的數據幀時內存分配的開銷。
如何使用groupby().transform
找到時間閾值,然后對整個數據使用布爾索引:
time_max_by_id = df.groupby('id')['time_entered'].transform('max') - pd.Timedelta('1D')
df[df['time_entered'] > time_max_by_id]
輸出:
id time_entered val
2 1 2015-02-24 18:00:00 0.978738
3 1 2015-02-25 03:00:00 2.240893
4 1 2015-02-25 12:00:00 1.867558
5 2 2015-02-25 21:00:00 -0.977278
6 2 2015-02-26 06:00:00 0.950088
11 3 2015-02-28 03:00:00 1.454274
12 3 2015-02-28 12:00:00 0.761038
13 3 2015-02-28 21:00:00 0.121675
df.groupby('id').apply(lambda x : x[(x['time_entered'].max()-x['time_entered'])<pd.Timedelta('1D')]).reset_index(drop=True)
Out[322]:
id time_entered val
0 1 2015-02-24 18:00:00 0.978738
1 1 2015-02-25 03:00:00 2.240893
2 1 2015-02-25 12:00:00 1.867558
3 2 2015-02-25 21:00:00 -0.977278
4 2 2015-02-26 06:00:00 0.950088
5 3 2015-02-28 03:00:00 1.454274
6 3 2015-02-28 12:00:00 0.761038
7 3 2015-02-28 21:00:00 0.121675
聲明:本站的技術帖子網頁,遵循CC BY-SA 4.0協議,如果您需要轉載,請注明本站網址或者原文地址。任何問題請咨詢:yoyou2525@163.com.