groupby中特定行的Python pandas差异

Question

我有一个熊猫数据框

df = pd.DataFrame({'Firm': ['Firm1','Firm1','Firm1','Firm1','Firm1','Firm1','Firm2','Firm2','Firm2','Firm2','Firm2','Firm2'],'Location' : ['Country1', 'Country1', 'Country1', 'Country2', 'Country2', 'Country2','Country1', 'Country1', 'Country1', 'Country2', 'Country2', 'Country2'], 'Currency' : ['Curr1', 'Curr2', 'Curr3', 'Curr1', 'Curr2', 'Curr3','Curr1', 'Curr2', 'Curr3', 'Curr1', 'Curr2', 'Curr3'], 'Value' : [100, 105, 110, 100, 95, 120, 95, 110, 115, 105, 120, 90] })

看起来像这样：

df:

     Firm  Location Currency  Value
0   Firm1  Country1    Curr1    100
1   Firm1  Country1    Curr2    105
2   Firm1  Country1    Curr3    110
3   Firm1  Country2    Curr1    100
4   Firm1  Country2    Curr2     95
5   Firm1  Country2    Curr3    120
6   Firm2  Country1    Curr1     95
7   Firm2  Country1    Curr2    110
8   Firm2  Country1    Curr3    115
9   Firm2  Country2    Curr1    105
10  Firm2  Country2    Curr2    120
11  Firm2  Country2    Curr3     90

现在我想计算每个公司位置组的 Curr3 和 Curr2 （列值）之间的差异，并根据结果更改 Curr3 的值。生成的 df 应如下所示：

     Firm  Location Currency  Value
0   Firm1  Country1    Curr1    100
1   Firm1  Country1    Curr2    105
2   Firm1  Country1    Curr3      5
3   Firm1  Country2    Curr1    100
4   Firm1  Country2    Curr2     95
5   Firm1  Country2    Curr3     25
6   Firm2  Country1    Curr1     95
7   Firm2  Country1    Curr2    110
8   Firm2  Country1    Curr3      5
9   Firm2  Country2    Curr1    105
10  Firm2  Country2    Curr2    120
11  Firm2  Country2    Curr3    -30

我尝试过使用.groupby 和.apply，这给了我结果，但是我想在原始数据框中进行转换。

df2 = df.groupby(['Firm','Location']).apply(lambda g: g[g.Currency == 'Curr3'].Value.values[0] - g[g.Currency == 'Curr2'].Value.values[0])

df2:

Firm    Location    0
Firm1   Country1    5
Firm1   Country2    25
Firm2   Country1    5
Firm2   Country2    -30

我无法弄清楚如何在原始 df 中就地执行此操作。我也使用.transform 进行了同样的尝试，但是它会产生错误：

df2 = df.groupby(['Firm','Location']).transform(lambda g: g[g.Currency == 'Curr3'].Value.values[0] - g[g.Currency == 'Curr2'].Value.values[0])

AttributeError: ("'Series' object has no attribute 'Currency'", 'occurred at index Currency')

----根据二凡的方案更新：

newvals = (
    df.where(df['Currency'].isin(['Curr2', 'Curr3']))
      .groupby(['Firm', 'Location'])['Value'].diff()
)
df['Value'] = newvals.fillna(df['Value'])

如果 df 看起来像这样（货币未排序），则解决方案不再有效（因为 diff() 仅计算与前一个值的差异

    Firm    Location    Currency    Value
0   Firm1   Country1    Curr2   100
1   Firm1   Country1    Curr1   105
2   Firm1   Country1    Curr3   110
3   Firm1   Country2    Curr3   100
4   Firm1   Country2    Curr2   95
5   Firm1   Country2    Curr1   120
6   Firm2   Country1    Curr1   95
7   Firm2   Country1    Curr2   110
8   Firm2   Country1    Curr3   115
9   Firm2   Country2    Curr2   105
10  Firm2   Country2    Curr3   120
11  Firm2   Country2    Curr1   90

-> 结果：

    Firm    Location    Currency    Value
0   Firm1   Country1    Curr2   100.0
1   Firm1   Country1    Curr1   105.0
2   Firm1   Country1    Curr3   10.0
3   Firm1   Country2    Curr3   100.0
4   Firm1   Country2    Curr2   -5.0
5   Firm1   Country2    Curr1   120.0
6   Firm2   Country1    Curr1   95.0
7   Firm2   Country1    Curr2   110.0
8   Firm2   Country1    Curr3   5.0
9   Firm2   Country2    Curr2   105.0
10  Firm2   Country2    Curr3   15.0
11  Firm2   Country2    Curr1   90.0

现在，不再每次计算 Curr3 和 Curr 2 之间的差值并替换 Curr3 的值。

Answer 1

使用DataFrame.where、Series.isin、GroupBy.diff 和Series.fillna：

首先我们将所有Curr1 转换为NaN 和where，然后我们对Firm 和Location 进行分组并计算Value 的差异。

newvals = (
    df.where(df['Currency'].isin(['Curr2', 'Curr3']))
      .groupby(['Firm', 'Location'])['Value'].diff()
)
df['Value'] = newvals.fillna(df['Value'])

     Firm  Location Currency  Value
0   Firm1  Country1    Curr1  100.0
1   Firm1  Country1    Curr2  105.0
2   Firm1  Country1    Curr3    5.0
3   Firm1  Country2    Curr1  100.0
4   Firm1  Country2    Curr2   95.0
5   Firm1  Country2    Curr3   25.0
6   Firm2  Country1    Curr1   95.0
7   Firm2  Country1    Curr2  110.0
8   Firm2  Country1    Curr3    5.0
9   Firm2  Country2    Curr1  105.0
10  Firm2  Country2    Curr2  120.0
11  Firm2  Country2    Curr3  -30.0