熊猫数据框中的并发通话分数之间的差异

时间:2019-07-02 16:54:12

标签: python pandas dataframe

我正在尝试使用用户@Garret提供的修改后的代码版本来分析以下数据集中的一些内容,但是我遇到了一些问题。

数据集的一列显示客户是被现场代理商还是自动机雇用。我正在尝试获取并发调用之间的区别,在该并发调用中,成员首先连接到代理,然后没有连接。该呼叫必须具有相同的呼叫原因,并且就时间戳而言,它必须放置在初始呼叫之后。此外,由于其他原因,在两者之间也可以打电话。

这是数据集:

data = [['bob13', 1, 'returns','automated',' 2019-08-18 10:12:00'],['bob13', 0, 'returns','automated',' 2019-03-18 10:12:00'],\
        ['bob13', 8, 'returns','agent',' 2019-04-18 10:15:00'],['rach2', 2, 'shipping','automated',' 2019-04-19 10:15:00'],\
        ['bob13', 0, 'returns','agent',' 2019-05-18 11:12:00'],['rach2', 0, 'shipping','agent',' 2019-04-18 11:15:00'],\
        ['bob13', 3, 'returns','agent',' 2019-02-18 10:12:00'],['rach2', 8, 'shipping','agent',' 2019-05-19 10:15:00'],\
       ['rach2', 7, 'shipping','automated',' 2019-06-19 10:15:00'],['roy', 4, 'exchange','agent','2019-03-26 17:36:00'],\
       ['roy', 5, 'exchange','automated','2019-01-28 09:48:00']]

df = pd.DataFrame(data, columns = ['member_id', 'survey_score','call_reason','connection','time_stamp']) 
df.sort_values(by=['time_stamp']).head(20)

member_id   survey_score    call_reason connection  time_stamp
6   bob13        3            returns   agent       2019-02-18 10:12:00
1   bob13        0            returns   automated   2019-03-18 10:12:00
2   bob13        8            returns   agent       2019-04-18 10:15:00
5   rach2        0            shipping  agent       2019-04-18 11:15:00
3   rach2        2            shipping  automated   2019-04-19 10:15:00
4   bob13        0            returns   agent       2019-05-18 11:12:00
7   rach2        8            shipping  agent       2019-05-19 10:15:00
8   rach2        7            shipping  automated   2019-06-19 10:15:00
0   bob13        1            returns   automated   2019-08-18 10:12:00
10  roy          5            exchange  automated   2019-01-28 09:48:00
9   roy          4            exchange  agent       2019-03-26 17:36:00




我期望的输出如下:

member_id    call_reason    automated    agent    score differential
bob13         returns           0          3            -3
bob13         returns           1          0             1
rach2         shipping          2          0             2
rach2         shipping          7          8            -1


因此,基本上,只需查找关于call_reason和connection的两个调用之间的区别。第一个呼叫是在成员连接到座席时,第二个呼叫必须基于时间戳而在第一个呼叫之后发出,出于相同的原因,并且必须连接到自动化系统。如果之间由于其他原因拨打了电话也可以。我尝试过的代码如下:

grp = df.query('connection=="automated"').\
    groupby(['member_id', 'call_reason'])
df['OutId'] = grp.time_stamp.transform(lambda x: x.rank())
df.head(10)
grp = df.groupby(['member_id', 'call_reason'])
df['Id'] = grp.OutId.transform(lambda x: x.bfill())
df.head(10)
agent = df.query('connection=="agent"').\
    groupby(['member_id', 'call_reason', 'Id']).survey_score.last()

automated = df.query('connection=="automated"').\
    groupby(['member_id', 'call_reason', 'Id']).survey_score.last()

ddf = pd.concat([automated, agent], axis=1,
                keys=['automated', 'agent'])
ddf['score_differential'] = ddf.automated - ddf.agent

我得到的输出是:

ddf.dropna().head(10)

                              automated     agent   score_differential
member_id   call_reason Id          
rach2         shipping  2.0      7           8.0          -1.0
roy           exchange  1.0      5           4.0           1.0



再次,预期输出将是:

member_id    call_reason    automated    agent    score differential
bob13         returns           0          3            -3
bob13         returns           1          0             1
rach2         shipping          2          0             2
rach2         shipping          7          8            -1

注意:如果解决方案可以灵活一些,我会很乐意,这样我就可以分析一些不同的情况,例如:

  1. 仅自动通话之间的区别
  2. 仅连接到业务代表的呼叫之间的区别
  3. 初始呼叫连接到座席时呼叫之间的差异,而第二个呼叫中的连接类型无关紧要

我们将不胜感激!

1 个答案:

答案 0 :(得分:2)

您可以通过创建一个函数,然后将该函数应用于groupby中的组来实现此目的。

设置初始数据框:

import pandas as pd

data = [['bob13', 1, 'returns','automated',' 2019-08-18 10:12:00'],['bob13', 0, 'returns','automated',' 2019-03-18 10:12:00'],\
        ['bob13', 8, 'returns','agent',' 2019-04-18 10:15:00'],['rach2', 2, 'shipping','automated',' 2019-04-19 10:15:00'],\
        ['bob13', 0, 'returns','agent',' 2019-05-18 11:12:00'],['rach2', 0, 'shipping','agent',' 2019-04-18 11:15:00'],\
        ['bob13', 3, 'returns','agent',' 2019-02-18 10:12:00'],['rach2', 8, 'shipping','agent',' 2019-05-19 10:15:00'],\
       ['rach2', 7, 'shipping','automated',' 2019-06-19 10:15:00'],['roy', 4, 'exchange','agent','2019-03-26 17:36:00'],\
       ['roy', 5, 'exchange','automated','2019-01-28 09:48:00']]

df = pd.DataFrame(data, columns = ['member_id', 'survey_score','call_reason','connection','time_stamp']) 
df.sort_values(by=['time_stamp']).head(20)
df['time_stamp'] = pd.to_datetime(df['time_stamp'])

df
   member_id  survey_score call_reason connection          time_stamp
0      bob13             1     returns  automated 2019-08-18 10:12:00
1      bob13             0     returns  automated 2019-03-18 10:12:00
2      bob13             8     returns      agent 2019-04-18 10:15:00
3      rach2             2    shipping  automated 2019-04-19 10:15:00
4      bob13             0     returns      agent 2019-05-18 11:12:00
5      rach2             0    shipping      agent 2019-04-18 11:15:00
6      bob13             3     returns      agent 2019-02-18 10:12:00
7      rach2             8    shipping      agent 2019-05-19 10:15:00
8      rach2             7    shipping  automated 2019-06-19 10:15:00
9        roy             4    exchange      agent 2019-03-26 17:36:00
10       roy             5    exchange  automated 2019-01-28 09:48:00

每当我尝试解决这样的问题时,我都会组成一个特定的小组。因此,我只是隔离了bob13,并尝试复制以达到我们想要的bob。导致我执行了一系列特定的步骤,然后将其放入函数中:

我们按时间对数据框进行排序,然后创建名为next_connection和'next_score'的新列。这些将结果从下一个结果中移出,以便我们将其放在该行中。我们删除所有丢失的内容(因为没有下一个,所以是该组的最后一个),我们隔离连接为agent且next_connection为automated的任何行。我们将列重命名以匹配您的输出,然后计算得分差异。

def function_(df):
    df = df.sort_values('time_stamp')
    df['next_connection'] = df.connection.shift(-1)
    df['next_score'] = df.survey_score.shift(-1)
    df = df.dropna()
    df = df[(df.connection == 'agent') & (df.next_connection == 'automated')]
    df = df.rename(columns={'survey_score':'agent', 'next_score':'automated'})
    df['score differential'] = df['automated'] - df['agent']
    return df

现在,我们将其应用于按member_idcall_reason分组的数据框。

g = df.groupby(['member_id', 'call_reason']).apply(function_)

g[['member_id','call_reason','automated','agent','score differential']].reset_index(drop=True)

  member_id call_reason  automated  agent  score differential
0     bob13     returns        0.0      3                -3.0
1     bob13     returns        1.0      0                 1.0
2     rach2    shipping        2.0      0                 2.0
3     rach2    shipping        7.0      8                -1.0