python - 重塑 Pandas 数据框失败

我想重塑我的数据框，它只有键值对。

例如，

             key                                              value
0     Message-ID       <5525962.1075855679785.JavaMail.evans@thyme>
1           Date              Wed, 13 Dec 2000 07:04:00 -0800 (PST)
2           From                            [e-mail]
3             To  [e-mail], james.steffes@enron...
4         X-From                                    Phillip K Allen
5           X-To  Christi L Nicolay, James D Steffes, Jeff Dasov...
6          X-cc:                                               None
7         X-bcc:                                               None
8       X-Origin                                            Allen-P
9     Message-ID       <4650921.1075855679981.JavaMail.evans@thyme>
10          Date               Tue, 5 Dec 2000 07:31:00 -0800 (PST)
11          From                               [e-mail]
12            To                             [e-mail]
13        X-From                                         Ina Rangel
14          X-To                                       Amanda Huble
15         X-cc:                                               None
16        X-bcc:                                               None
17      X-Origin                                            Allen-P

我想将其转换为：

Message-ID       Date                  From             To        X-From                 X-To                            X-cc:  X-bcc:  X-Origin
<5525962.10...   Wed, 13 Dec 2000...   phillip.allen... christi.nicolay.. Phillip K Allen..     Christi L Nicolay, Ja... NaN    NaN     Allen-P
<4650921.10...   Tue, 5 Dec 2000 ...   ina.rangel...    amanda.huble@...  Ina Rangel            Amanda Huble             NaN    NaN     Allen-P

我尝试过旋转表，但是我很困惑应该作为索引参数的内容。请帮我解决这个问题。

如果发现重复，请随意将其标记为重复。

最佳答案

如果每个组总是有9个值，则对numpy.reshape和2d array构造函数使用DataFrame对于key，对于列值，也应采用列Message-ID的前9个值：

print (df['value'].values.reshape(-1, 9))
[['<5525962.1075855679785.JavaMail.evans@thyme>'
  'Wed, 13 Dec 2000 07:04:00 -0800 (PST)' '[e-mail]'
  '[e-mail], james.steffes@enron...' 'Phillip K Allen'
  'Christi L Nicolay, James D Steffes, Jeff Dasov...' 'None' 'None'
  'Allen-P']
 ['<4650921.1075855679981.JavaMail.evans@thyme>'
  'Tue, 5 Dec 2000 07:31:00 -0800 (PST)' '[e-mail]'
  '[e-mail]' 'Ina Rangel' 'Amanda Huble' 'None' 'None'
  'Allen-P']]


df = pd.DataFrame(df['value'].values.reshape(-1, 9), columns=df['key'].iloc[:9])
print (df)
key                                    Message-ID  \
0    <5525962.1075855679785.JavaMail.evans@thyme>
1    <4650921.1075855679981.JavaMail.evans@thyme>

key                                   Date                     From  \
0    Wed, 13 Dec 2000 07:04:00 -0800 (PST)  [e-mail]
1     Tue, 5 Dec 2000 07:31:00 -0800 (PST)     [e-mail]

key                                                 To           X-From  \
0    [e-mail], james.steffes@enron...  Phillip K Allen
1                               [e-mail]       Ina Rangel

key                                               X-To X-cc: X-bcc: X-Origin
0    Christi L Nicolay, James D Steffes, Jeff Dasov...  None   None  Allen-P
1                                         Amanda Huble  None   None  Allen-P

如果总是可能在每个组中的数据中出现set_index行，则将Series与由布尔掩码cumsum创建的助手eq一起使用-与== 进行比较以标识每个组的开始：

df = df.set_index([df['key'].eq('Message-ID').cumsum(), 'key'])['value'].unstack()
print (df)
key                                   Date                     From  \
key
1    Wed, 13 Dec 2000 07:04:00 -0800 (PST)  [e-mail]
2     Tue, 5 Dec 2000 07:31:00 -0800 (PST)     [e-mail]

key                                    Message-ID  \
key
1    <5525962.1075855679785.JavaMail.evans@thyme>
2    <4650921.1075855679981.JavaMail.evans@thyme>

key                                                 To           X-From  \
key
1    [e-mail], james.steffes@enron...  Phillip K Allen
2                               [e-mail]       Ina Rangel

key X-Origin                                               X-To X-bcc: X-cc:
key
1    Allen-P  Christi L Nicolay, James D Steffes, Jeff Dasov...   None  None
2    Allen-P                                       Amanda Huble   None  None