根據列的總和添加新列，並在Pandas中按2個其他列分組

Question

我有數據幀：

df = pd.DataFrame({'Continent':['North America','North America','North America','Europe','Europe','Europe','Europe'],
                'Country': ['US','Canada','Mexico','France','Germany','Spain','Italy'],
                'Status': ['Member','Non-Member','Non-Member','Member','Non-Member','Member','Non-Member'],
                'Units': [27,5,4,10,15,8,8]})

print df

       Continent  Country      Status  Units
0  North America       US      Member     27
1  North America   Canada  Non-Member      5
2  North America   Mexico  Non-Member      4
3         Europe   France      Member     10
4         Europe  Germany  Non-Member     15
5         Europe    Spain      Member      8
6         Europe    Italy  Non-Member      8

我需要添加2列，這些列是關於大陸的摘要統計信息。 我需要一個列有成員國和非成員國單位總和的專欄。

這樣最終輸出看起來像：

       Continent  Member Units  Non-Member Units  Country      Status  Units
0  North America            27                 9       US      Member     27
1  North America            27                 9   Canada  Non-Member      5
2  North America            27                 9   Mexico  Non-Member      4
3         Europe            18                23   France      Member     10
4         Europe            18                23  Germany  Non-Member     15
5         Europe            18                23    Spain      Member      8
6         Europe            18                23    Italy  Non-Member      8

看起來我需要使用groupby，但我無法弄清楚如何獲取groupby值並將它們作為新列重新插入數據幀。

summary_stats = df.groupby(['Continent','Status'])['Units'].sum()
print summary_stats

Continent      Status    
Europe         Member        18
               Non-Member    23
North America  Member        27
               Non-Member     9
Name: Units, dtype: int64

我也嘗試過不使用groupby：

df['Member Units'] = df['Units'][df['Status'] == 'Member'].sum()
df['Non-Member Units'] = df['Units'][df['Status'] == 'Non-Member'].sum()

但這並沒有因為大陸而有所區別，所以它只是將所有會員和非會員加起來

任何幫助都非常有用！

Answer 1

我認為你需要第一個groupby並transform sum來創建新的all_sum Series 。 然后我認為最好使用numpy.where並且如果是成員，從Series獲取值，如果不是，則獲得0 。 與非成員類似：

all_sum = df.groupby(['Continent','Status'])['Units'].transform(sum)
print all_sum
0    27
1     9
2     9
3    18
4    23
5    18
6    23
dtype: int64

df['Member Units'] = np.where(df['Status'] == 'Member', all_sum, 0)
df['Non-Member Units'] = np.where(df['Status'] != 'Member', all_sum, 0)
print df
       Continent  Country      Status  Units  Member Units  Non-Member Units
0  North America       US      Member     27            27                 0
1  North America   Canada  Non-Member      5             0                 9
2  North America   Mexico  Non-Member      4             0                 9
3         Europe   France      Member     10            18                 0
4         Europe  Germany  Non-Member     15             0                23
5         Europe    Spain      Member      8            18                 0
6         Europe    Italy  Non-Member      8             0                23

Answer 2

一旦你有summary_stats我認為你可以這樣做：

df['Member Units'] = summary_stats[zip(df['Continent'].values, df['Status'].values)]

您需要zip Series值的原因是df['Continent']返回帶索引的系列，但您不希望這種情況發生。

Answer 3

既然你有summary_stats ，你可以在重塑它之后使用merge() ：

summary = summary_stats.reset_index().pivot(index='Continent', columns='Status', values='Units')

summary['Continent'] = summary.index

df = df.merge(summary, on='Continent')

然后只需根據需要重命名列