Python 数据帧计算

我有一个相当复杂的数据框架，看起来像这样：

df = pd.DataFrame({'0': {('Total Number of End Points', '0.01um', '0hr'): 12,
  ('Total Number of End Points', '0.1um', '0hr'): 8,
  ('Total Number of End Points', 'Control', '0hr'): 4,
  ('Total Number of End Points', '0.01um', '24hr'): 18,
  ('Total Number of End Points', '0.1um', '24hr'): 12,
  ('Total Number of End Points', 'Control', '24hr'): 6,
  ('Total Vessel Length', '0.01um', '0hr'): 12,
  ('Total Vessel Length', '0.1um', '0hr'): 8,
  ('Total Vessel Length', 'Control', '0hr'): 4,
  ('Total Vessel Length', '0.01um', '24hr'): 18,
  ('Total Vessel Length', '0.1um',  '24hr'): 12,
  ('Total Vessel Length', 'Control',  '24hr'): 6},
  '1': {('Total Number of End Points', '0.01um', '0hr'): 12,
  ('Total Number of End Points', '0.1um', '0hr'): 8,
  ('Total Number of End Points', 'Control', '0hr'): 4,
  ('Total Number of End Points', '0.01um', '24hr'): 18,
  ('Total Number of End Points', '0.1um', '24hr'): 12,
  ('Total Number of End Points', 'Control', '24hr'): 6,
  ('Total Vessel Length', '0.01um', '0hr'): 12,
  ('Total Vessel Length', '0.1um', '0hr'): 8,
  ('Total Vessel Length', 'Control', '0hr'): 4,
  ('Total Vessel Length', '0.01um', '24hr'): 18,
  ('Total Vessel Length', '0.1um',  '24hr'): 12,
  ('Total Vessel Length', 'Control',  '24hr'): 6},
  '2': {('Total Number of End Points', '0.01um', '0hr'): 12,
  ('Total Number of End Points', '0.1um', '0hr'): 8,
  ('Total Number of End Points', 'Control', '0hr'): 4,
  ('Total Number of End Points', '0.01um', '24hr'): 18,
  ('Total Number of End Points', '0.1um', '24hr'): 12,
  ('Total Number of End Points', 'Control', '24hr'): 6,
  ('Total Vessel Length', '0.01um', '0hr'): 12,
  ('Total Vessel Length', '0.1um', '0hr'): 8,
  ('Total Vessel Length', 'Control', '0hr'): 4,
  ('Total Vessel Length', '0.01um', '24hr'): 18,
  ('Total Vessel Length', '0.1um',  '24hr'): 12,
  ('Total Vessel Length', 'Control',  '24hr'): 6}})

print(df)
                                                 0   1   2
        Total Number of End Points 0.01um  0hr   12  12  12
                                           24hr  18  18  18
                                   0.1um   0hr    8   8   8
                                           24hr  12  12  12
                                   Control 0hr    4   4   4
                                           24hr   6   6   6
        Total Vessel Length        0.01um  0hr   12  12  12
                                           24hr  18  18  18
                                   0.1um   0hr    8   8   8
                                           24hr  12  12  12
                                   Control 0hr    4   4   4
                                           24hr   6   6   6

我试图将每个值除以相应控制级别中列的平均值。我尝试了以下方法，但没有成功

df2 = df.divide(df.xs('Control', level=1).mean(axis=1), axis='index')

我对python和pandas非常陌生，所以我倾向于用MS Excel的术语来思考这个问题

如果在Excel中，A1的公式（‘终点总数’、‘0.01um’、‘0hr’、‘0）看起来是：

=A1/平均值（$A$5:$C$5）

B1（‘终点总数’、‘0.01um’、‘0hr’、1）将为：

=B1/平均值（$A$5:$C$5）

A2（‘终点总数’、‘0.01um’、‘24小时’、‘0’）为

=A1/平均值（$A$6:$C$6）

本例的预期结果为：

                                                 0  1  2
        Total Number of End Points 0.01um  0hr   3  3  3
                                           24hr  3  3  3
                                   0.1um   0hr   2  2  2
                                           24hr  2  2  2
                                   Control 0hr   1  1  1
                                           24hr  1  1  1
        Total Vessel Length        0.01um  0hr   3  3  3
                                           24hr  3  3  3
                                   0.1um   0hr   2  2  2
                                           24hr  2  2  2
                                   Control 0hr   1  1  1
                                           24hr  1  1  1

---

注：实际数据中有许多索引和列。

这里的问题是，熊猫的组织方式很容易计算列，问题是需要从其他行中减去一行的平均值。熊猫不是设计用来工作的

但是，您可以使用transpose

.T

轻松地切换行和列，这样它可能更容易处理，事实上，控制平均值是一行

>>> df.T[(u'Total Vessel Length', u'Control', u'0hr')].mean()
4.0

该4.0来自原始数据中的两个4.0值：

>>> df.T[(u'Total Vessel Length', u'Control', u'0hr')]
a    4
b    4

现在看来for循环将解决这个问题

未经测试：

for primary in (u'Total Vessel Length',u'Total Number of End Points'):
     for um in (u'0.01um',u'0.1um'):
         for hours in (u'0hr',u'24hr'):
             df.T[(primary,um,hours)]=df.T[(primary,um,hours)]/df.T[(primary, u'Control', hours)].mean()

请注意，这并不划分非控制列，但很容易将“控制”包含到um循环中

更新这不起作用，不知何故它没有在适当的位置修改数据帧。现在，我不知道为什么

但是您可以通过在dict上调用pd.DataFrame来构造新的数据帧 理解力

这似乎起作用了

import pandas as pd

df = pd.DataFrame({'0': {('Total Number of End Points', '0.01um', '0hr'): 12,
  ('Total Number of End Points', '0.1um', '0hr'): 8,
  ('Total Number of End Points', 'Control', '0hr'): 4,
  ('Total Number of End Points', '0.01um', '24hr'): 18,
  ('Total Number of End Points', '0.1um', '24hr'): 12,
  ('Total Number of End Points', 'Control', '24hr'): 6,
  ('Total Vessel Length', '0.01um', '0hr'): 12,
  ('Total Vessel Length', '0.1um', '0hr'): 8,
  ('Total Vessel Length', 'Control', '0hr'): 4,
  ('Total Vessel Length', '0.01um', '24hr'): 18,
  ('Total Vessel Length', '0.1um',  '24hr'): 12,
  ('Total Vessel Length', 'Control',  '24hr'): 6},
  '1': {('Total Number of End Points', '0.01um', '0hr'): 12,
  ('Total Number of End Points', '0.1um', '0hr'): 8,
  ('Total Number of End Points', 'Control', '0hr'): 4,
  ('Total Number of End Points', '0.01um', '24hr'): 18,
  ('Total Number of End Points', '0.1um', '24hr'): 12,
  ('Total Number of End Points', 'Control', '24hr'): 6,
  ('Total Vessel Length', '0.01um', '0hr'): 12,
  ('Total Vessel Length', '0.1um', '0hr'): 8,
  ('Total Vessel Length', 'Control', '0hr'): 4,
  ('Total Vessel Length', '0.01um', '24hr'): 18,
  ('Total Vessel Length', '0.1um',  '24hr'): 12,
  ('Total Vessel Length', 'Control',  '24hr'): 6},
  '2': {('Total Number of End Points', '0.01um', '0hr'): 12,
  ('Total Number of End Points', '0.1um', '0hr'): 8,
  ('Total Number of End Points', 'Control', '0hr'): 4,
  ('Total Number of End Points', '0.01um', '24hr'): 18,
  ('Total Number of End Points', '0.1um', '24hr'): 12,
  ('Total Number of End Points', 'Control', '24hr'): 6,
  ('Total Vessel Length', '0.01um', '0hr'): 12,
  ('Total Vessel Length', '0.1um', '0hr'): 8,
  ('Total Vessel Length', 'Control', '0hr'): 4,
  ('Total Vessel Length', '0.01um', '24hr'): 18,
  ('Total Vessel Length', '0.1um',  '24hr'): 12,
  ('Total Vessel Length', 'Control',  '24hr'): 6}})

print df

df2 = pd.DataFrame({(primary,um,hours):df.T[(primary,um,hours)]/df.T[(primary,u'Control',hours)].mean() for primary in (u'Total Vessel Length',u'Total Number of End Points') for um in (u'0.01um',u'0.1um') for hours in (u'0hr',u'24hr')})

print df2.T

输出

paul@home:~/SO$ python ./r.py 
                                              0   1   2
(Total Number of End Points, 0.01um, 0hr)    12  12  12
(Total Number of End Points, 0.01um, 24hr)   18  18  18
(Total Number of End Points, 0.1um, 0hr)      8   8   8
(Total Number of End Points, 0.1um, 24hr)    12  12  12
(Total Number of End Points, Control, 0hr)    4   4   4
(Total Number of End Points, Control, 24hr)   6   6   6
(Total Vessel Length, 0.01um, 0hr)           12  12  12
(Total Vessel Length, 0.01um, 24hr)          18  18  18
(Total Vessel Length, 0.1um, 0hr)             8   8   8
(Total Vessel Length, 0.1um, 24hr)           12  12  12
(Total Vessel Length, Control, 0hr)           4   4   4
(Total Vessel Length, Control, 24hr)          6   6   6

[12 rows x 3 columns]
                                            0  1  2
(Total Number of End Points, 0.01um, 0hr)   3  3  3
(Total Number of End Points, 0.01um, 24hr)  3  3  3
(Total Number of End Points, 0.1um, 0hr)    2  2  2
(Total Number of End Points, 0.1um, 24hr)   2  2  2
(Total Vessel Length, 0.01um, 0hr)          3  3  3
(Total Vessel Length, 0.01um, 24hr)         3  3  3
(Total Vessel Length, 0.1um, 0hr)           2  2  2
(Total Vessel Length, 0.1um, 24hr)          2  2  2

[8 rows x 3 columns]

---

将

控件

值放在它们自己的列中会有所帮助。您可以使用

取消堆叠
：

df.index.names = ['field', 'type', 'time']
df2 = df.unstack(['type']).swaplevel(0, 1, axis=1)

# type                            0.01um 0.1um Control 0.01um 0.1um Control  \
#                                      0     0       0      1     1       1   
# field                      time                                             
# Total Number of End Points 0hr      12     8       4     12     8       4   
#                            24hr     18    12       6     18    12       6   
# Total Vessel Length        0hr      12     8       4     12     8       4   
#                            24hr     18    12       6     18    12       6   

# type                            0.01um 0.1um Control  
#                                      2     2       2  
# field                      time                       
# Total Number of End Points 0hr      12     8       4  
#                            24hr     18    12       6  
# Total Vessel Length        0hr      12     8       4  
#                            24hr     18    12       6  

现在找到每个控件的平均值：

ave = df2['Control'].mean(axis=1)
# field                       time
# Total Number of End Points  0hr     4
#                             24hr    6
# Total Vessel Length         0hr     4
#                             24hr    6
# dtype: float64

正如您所期望的，您可以使用
df2.divide
来计算所需的结果。确保使用
axis=0
告诉熊猫根据行索引匹配值（在
df2
和
ave
）

result = df2.divide(ave, axis=0)
# type                            0.01um 0.1um Control 0.01um 0.1um Control  \
#                                      0     0       0      1     1       1   
# field                      time                                             
# Total Number of End Points 0hr       3     2       1      3     2       1   
#                            24hr      3     2       1      3     2       1   
# Total Vessel Length        0hr       3     2       1      3     2       1   
#                            24hr      3     2       1      3     2       1   

# type                            0.01um 0.1um Control  
#                                      2     2       2  
# field                      time                       
# Total Number of End Points 0hr       3     2       1  
#                            24hr      3     2       1  
# Total Vessel Length        0hr       3     2       1  
#                            24hr      3     2       1  

本质上，你追求的是价值观。但是，如果您希望重新排列数据帧，使其与您发布的内容完全一致，则：

result = result.stack(['type'])
result = result.reorder_levels(['field','type','time'], axis=0)
result = result.reindex(df.index)

屈服

                                         0  1  2
field                      type    time         
Total Number of End Points 0.01um  0hr   3  3  3
                                   24hr  3  3  3
                           0.1um   0hr   2  2  2
                                   24hr  2  2  2
                           Control 0hr   1  1  1
                                   24hr  1  1  1
Total Vessel Length        0.01um  0hr   3  3  3
                                   24hr  3  3  3
                           0.1um   0hr   2  2  2
                                   24hr  2  2  2
                           Control 0hr   1  1  1
                                   24hr  1  1  1


总而言之：

df.index.names = ['field', 'type', 'time']
df2 = df.unstack(['type']).swaplevel(0, 1, axis=1)
ave = df2['Control'].mean(axis=1)
result = df2.divide(ave, axis=0)
result = result.stack(['type'])
result = result.reorder_levels(['field','type','time'], axis=0)
result = result.reindex(df.index)

你能提供一个期望输出的例子吗？当我把你的问题顶部的数据放入一个数据框时，它与你用print（df）得到的不同。df=。。。打印（df）是两个不同的数据帧。您的打印（df）与上述代码无关。您的输入列为['a'，'b'，，但打印列为[0,1,2]。你能让一切保持一致吗。谢谢。@MarkGraph哇塞。。你说得对。。我会解决的。在熊猫中，数据在内部是按列组织的，因此提取或计算列是最容易的。您是否可以重新组织数据，使所有控件值都位于它们自己的列中？我得到的结果与中相同。在什么地方需要
inplace=True
吗？这里也一样。看起来非常熟悉。我会四处看看，也许是相关的。还在看。有趣。我没有注意到索引可以是元组，并且有所有这些关联的方法。