具有多个条件和列的Pandas Groupby和cumsum-Python

编程问答 2022-06-06

问题描述

我有以下数据框：

import pandas as pd

hits = {'id': ['A','A','B','C','C'],'datetime': ['2010-01-02 03:00:00','2010-01-02 03:00:14','2010-01-02 03:00:35','2010-01-02 03:00:38','2010-01-02 03:29:10','2010-01-02 03:29:35','2010-01-02 03:45:20','2010-01-02 06:10:05','2010-01-02 06:10:15','2010-01-02 07:40:15','2010-01-02 07:40:20','2010-01-02 07:40:25'],'uri_len': [10,20,25,15,10,30,40,45]
       }

df = pd.DataFrame(hits,columns = ['id','datetime','uri_len'])

df['datetime'] =  pd.to_datetime(df['datetime'],format='%Y-%m-%d %H:%M:%s')

print (df)

   id            datetime  uri_len
0   A 2010-01-02 03:00:00       10
1   A 2010-01-02 03:00:14       20
2   A 2010-01-02 03:00:35       25
3   A 2010-01-02 03:00:38       15
4   A 2010-01-02 03:29:10       20
5   A 2010-01-02 03:29:35       10
6   B 2010-01-02 03:45:20       20
7   B 2010-01-02 06:10:05       25
8   B 2010-01-02 06:10:15       15
9   C 2010-01-02 07:40:15       30
10  C 2010-01-02 07:40:20       40
11  C 2010-01-02 07:40:25       45

我想使用id作为变量分组按会话对命中进行分组。对我来说，会话​​的闲置时间超过15秒（从datetime列计算），或者uri_len列减少，并且在两种情况下都比较连续命中。

我知道如何分别根据每个条件进行分组：

df['session1'] = (df.groupby('id')['datetime']
               .transform(lambda x: x.diff().gt('15Sec').cumsum())
              )

df['session2'] = (df.groupby('id')['uri_len']
               .transform(lambda x: x.diff().lt(0).cumsum())
              )

有没有办法在同一行中组合两个转换，所以输出就是这个？：

   id            datetime  uri_len  session
0   A 2010-01-02 03:00:00       10        0
1   A 2010-01-02 03:00:14       20        0
2   A 2010-01-02 03:00:35       25        1
3   A 2010-01-02 03:00:38       15        2
4   A 2010-01-02 03:29:10       20        3
5   A 2010-01-02 03:29:35       10        4
6   B 2010-01-02 03:45:20       20        0
7   B 2010-01-02 06:10:05       25        1
8   B 2010-01-02 06:10:15       15        2
9   C 2010-01-02 07:40:15       30        0
10  C 2010-01-02 07:40:20       40        0
11  C 2010-01-02 07:40:25       45        0

解决方法

如果我理解正确，您想添加它们吗？

import matplotlib.pyplot as plt
import numpy as np
import matplotlib.animation as animation

n = 100        
x = np.random.randn(n)


def update(curr):
    if curr == n: 
        a.event_source.stop()

plt.cla()
bins = np.arange(-4,4,0.5)
plt.hist(x[:curr],bins=bins)
plt.axis([-4,30])
plt.gca().set_title('Sampling the Normal Distribution')
plt.gca().set_ylabel('Frequency')
plt.gca().set_xlabel('Value')
plt.annotate('n = {}'.format(curr),[3,27])

fig = plt.figure()         
a = animation.FuncAnimation(fig,update,interval=10)

更清晰的方法：

df['session'] = df.groupby('id')['datetime'].transform(lambda x: 
x.diff().gt('15Sec').cumsum()) + df.groupby('id')['uri_len'].transform(lambda x: 
x.diff().lt(0).cumsum())
cumsumpandaspandas-groupbypython