Python数据帧整形

我有一个python数据框架，可以简化如下：

python
df= pd.DataFrame([['January','Monday',np.nan,np.nan,np.nan,1,20],['January','Monday',np.nan,np.nan,np.nan,2,25],['February','Monday',np.nan,np.nan,np.nan,1,15],\
      ['February','Monday',np.nan,np.nan,np.nan,2,20],['February','Monday',np.nan,np.nan,np.nan,3,25],['March','Tuesday',np.nan,np.nan,np.nan,1,50],\
      ['March','Wednesday',np.nan,np.nan,np.nan,1,75]],columns = ['Month','Day','Data1','Data2', 'Data3','Count','Initial_Data'])

新数据框架的目的/目标：我希望按月份和日期对数据进行分类。我想用来自初始_数据的图填充列Data1、Data2和Data3。例如，对于一月和星期一，Data1=20，Data2=25，Data3保持为NaN，因为一月和星期一的计数最高=2。对于二月和星期一，我希望数据1=15，数据2=20和数据3=25，这是因为二月和星期一的计数最高=3。对于三月份的星期二，我希望Data1=50，Data2和Data3=NaN，对于三月份的星期三，我希望Data1=75和Data2=Data3=NaN，因为它们的最高计数为1。最终数据将如下所示：

      Month        Day  Data1  Data2  Data3
0   January     Monday     20   25.0    NaN
1   January     Monday     20   25.0    NaN
2  February     Monday     15   20.0   25.0
3  February     Monday     15   20.0   25.0
4  February     Monday     15   20.0   25.0
5     March    Tuesday     50    NaN    NaN
6     March  Wednesday     75    NaN    NaN

我尝试使用if语句，但它不起作用，因为我找不到填充所有三列（Data1、Data2和Data3）的解决方案。非常感谢。

您可以尝试以下方法：

df2 = df.set_index(['Month','Day','Count'])['Initial_Data'].unstack().add_prefix('Data').reset_index()
df.merge(df2, on=['Month','Day'], suffixes=('_x',''))[df.columns]

输出：

      Month        Day  Data1  Data2  Data3  Count  Initial_Data
0   January     Monday   20.0   25.0    NaN      1            20
1   January     Monday   20.0   25.0    NaN      2            25
2  February     Monday   15.0   20.0   25.0      1            15
3  February     Monday   15.0   20.0   25.0      2            20
4  February     Monday   15.0   20.0   25.0      3            25
5     March    Tuesday   50.0    NaN    NaN      1            50
6     March  Wednesday   75.0    NaN    NaN      1            75

详情：

首先，使用

set_index

和

unstack

最里面的索引移动到to列中的“Count”。从而重塑数据帧。然后将“数据”前缀添加到列标题

---

接下来，我们需要

合并

或根据月和日列将两个数据框连接在一起。

这是我的答案，但Scott用了一个更好的答案击败了我

import numpy as np
import pandas as pd

df = pd.DataFrame([
    ['January','Monday',np.nan,np.nan,np.nan,1,20],\
    ['January','Monday',np.nan,np.nan,np.nan,2,25],\
    ['February','Monday',np.nan,np.nan,np.nan,1,15],\
    ['February','Monday',np.nan,np.nan,np.nan,2,20],\
    ['February','Monday',np.nan,np.nan,np.nan,3,25],\
    ['March','Tuesday',np.nan,np.nan,np.nan,1,50],\
    ['March','Wednesday',np.nan,np.nan,np.nan,1,75]],
    columns = ['Month','Day','Data1','Data2', 'Data3','Count','Initial_Data'])

new = pd.DataFrame(columns = ['Month','Day','Data1','Data2', 'Data3'])

for ridx, row in df.iterrows():
    new.loc[ridx] = [row['Month'], row['Day'], np.nan, np.nan, np.nan]
    if row['Count'] == 1:
        new.loc[new.index[ridx], 'Data1'] = row['Initial_Data']
    if row['Count'] == 2:
        new.loc[new.index[ridx], 'Data2'] = row['Initial_Data']
        new.loc[new.index[ridx-1], 'Data2'] = row['Initial_Data']
        new.loc[new.index[ridx], 'Data1'] = new.loc[new.index[ridx-1], 'Data1']
    if row['Count'] == 3:
        new.loc[new.index[ridx], 'Data3'] = row['Initial_Data']
        new.loc[new.index[ridx-1], 'Data3'] = row['Initial_Data']
        new.loc[new.index[ridx-2], 'Data3'] = row['Initial_Data']
        new.loc[new.index[ridx], 'Data1'] = new.loc[new.index[ridx-1], 'Data1']
        new.loc[new.index[ridx], 'Data2'] = new.loc[new.index[ridx-1], 'Data2']

print(new)

---

---

你能再解释一下这到底是怎么回事吗？@Ryan你最好把这句话一部分一部分地说出来，看看是怎么回事。如果你不是一个数据库人员，最难理解的部分是创建笛卡尔积的概念。使用

merge

方法在重复键上创建多个。我可以阅读/研究一下笛卡尔积的含义吗？我已经一步一步地运行了它，但这是我不太明白的路线。试试看

import numpy as np
import pandas as pd

df = pd.DataFrame([
    ['January','Monday',np.nan,np.nan,np.nan,1,20],\
    ['January','Monday',np.nan,np.nan,np.nan,2,25],\
    ['February','Monday',np.nan,np.nan,np.nan,1,15],\
    ['February','Monday',np.nan,np.nan,np.nan,2,20],\
    ['February','Monday',np.nan,np.nan,np.nan,3,25],\
    ['March','Tuesday',np.nan,np.nan,np.nan,1,50],\
    ['March','Wednesday',np.nan,np.nan,np.nan,1,75]],
    columns = ['Month','Day','Data1','Data2', 'Data3','Count','Initial_Data'])

new = pd.DataFrame(columns = ['Month','Day','Data1','Data2', 'Data3'])

for ridx, row in df.iterrows():
    new.loc[ridx] = [row['Month'], row['Day'], np.nan, np.nan, np.nan]
    if row['Count'] == 1:
        new.loc[new.index[ridx], 'Data1'] = row['Initial_Data']
    if row['Count'] == 2:
        new.loc[new.index[ridx], 'Data2'] = row['Initial_Data']
        new.loc[new.index[ridx-1], 'Data2'] = row['Initial_Data']
        new.loc[new.index[ridx], 'Data1'] = new.loc[new.index[ridx-1], 'Data1']
    if row['Count'] == 3:
        new.loc[new.index[ridx], 'Data3'] = row['Initial_Data']
        new.loc[new.index[ridx-1], 'Data3'] = row['Initial_Data']
        new.loc[new.index[ridx-2], 'Data3'] = row['Initial_Data']
        new.loc[new.index[ridx], 'Data1'] = new.loc[new.index[ridx-1], 'Data1']
        new.loc[new.index[ridx], 'Data2'] = new.loc[new.index[ridx-1], 'Data2']

print(new)

      Month        Day Data1 Data2 Data3
0   January     Monday    20    25   NaN
1   January     Monday    20    25   NaN
2  February     Monday    15    20    25
3  February     Monday    15    20    25
4  February     Monday    15    20    25
5     March    Tuesday    50   NaN   NaN
6     March  Wednesday    75   NaN   NaN