pandas - 保留顺序时的数据透视表失败

Question

我有以下数据框，其中周不是 ISO 周而是财政周（1 是 7 月的第一周，52 是 6 月的最后一周）：

> df
     domain  week  count
0        A    43      5
1        A    45      1
2        A    50      1
3        A    51      4
4        A     1      3
5        A     3     12
6        B    43      1
7        B    44      1
8        B    45      4
9        B    50     11
10       B     2      3
11       B     3     12
12       C    51      6
13       C     1     14
14       C     5      1

我想在保留星期顺序的同时旋转该表，以获得一个新的数据框，该数据框如下所示，其中值为计数，列为域：

> new_df
week   A      B     C
43      5     1   NaN
44    NaN     1   NaN
45      1     4   NaN      
50      1    11   NaN
51      4   NaN     6
1       3   NaN    14
2     NaN     3   NaN
3      12    12   NaN
5     NaN   NaN     1

我尝试使用 groupie 并解开，但收到此错误：

> df = df.groupby(['week'], sort=False)['count'].unstack('domain')
AttributeError: Cannot access callable attribute 'unstack' of 'SeriesGroupBy' objects, try using the 'apply' method

Answer 1

选项 1] 您可以使用自定义排序的weeks 索引助手和.loc

In [4810]: weeks = pd.Index(list(range(26, 52)) + list(range(26)))

In [4819]: dfp = df.groupby(['week','domain'])['count'].sum().unstack()

In [4820]: dfp.loc[weeks & dfp.index]
Out[4820]:
domain     A     B     C
43       5.0   1.0   NaN
44       NaN   1.0   NaN
45       1.0   4.0   NaN
50       1.0  11.0   NaN
51       4.0   NaN   6.0
1        3.0   NaN  14.0
2        NaN   3.0   NaN
3       12.0  12.0   NaN
5        NaN   NaN   1.0

选项 2] 或者，使用 pivot

In [4821]: dfp = df.pivot('week', 'domain', 'count')

In [4822]: dfp.loc[weeks & dfp.index]
Out[4822]:
domain     A     B     C
43       5.0   1.0   NaN
44       NaN   1.0   NaN
45       1.0   4.0   NaN
50       1.0  11.0   NaN
51       4.0   NaN   6.0
1        3.0   NaN  14.0
2        NaN   3.0   NaN
3       12.0  12.0   NaN
5        NaN   NaN   1.0

选项 3] 或者，reindex 而不是 .loc

In [4830]: dfp.reindex(weeks & dfp.index)
Out[4830]:
domain     A     B     C
43       5.0   1.0   NaN
44       NaN   1.0   NaN
45       1.0   4.0   NaN
50       1.0  11.0   NaN
51       4.0   NaN   6.0
1        3.0   NaN  14.0
2        NaN   3.0   NaN
3       12.0  12.0   NaN
5        NaN   NaN   1.0

---

详情

In [4826]: weeks
Out[4826]:
Int64Index([26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42,
            43, 44, 45, 46, 47, 48, 49, 50, 51,  0,  1,  2,  3,  4,  5,  6,  7,
             8,  9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24,
            25],
           dtype='int64')

In [4827]: weeks & dfp.index
Out[4827]: Int64Index([43, 44, 45, 50, 51, 1, 2, 3, 5], dtype='int64')

Answer 2

你需要weeks的自定义顺序，所以需要ordered categorical自定义顺序并省略sort=False：

cats = list(range(26, 52)) + list(range(26))
print (cats)
[26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 
 47, 48, 49, 50, 51, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 
 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]

df['week'] = df['week'].astype('category', ordered=True, categories=cats)

df = df.groupby(['week','domain'])['count'].sum().unstack()
print (df)
domain     A     B     C
week                    
43       5.0   1.0   NaN
44       NaN   1.0   NaN
45       1.0   4.0   NaN
50       1.0  11.0   NaN
51       4.0   NaN   6.0
1        3.0   NaN  14.0
2        NaN   3.0   NaN
3       12.0  12.0   NaN
5        NaN   NaN   1.0