pandas-groupby子组的频率计算，新行的插入和列的重新排列

Question

我需要一些帮助才能对子组执行一些操作，但我真的很困惑。我将尝试使用 cmets 快速描述操作和所需的输出。

(1) 计算每个子组的出现频率百分比

(2) 出现一条不存在的记录，带0

(3) 重新排列记录和列的顺序

假设下面的df为原始数据：

df=pd.DataFrame({'store':[1,1,1,2,2,2,3,3,3,3],
                 'branch':['A','A','C','C','C','C','A','A','C','A'],
                 'products':['clothes', 'shoes', 'clothes', 'shoes', 'accessories', 'clothes', 'bags', 'bags', 'clothes', 'clothes']})

下面的 grouped_df 与我的想法很接近，但我无法获得所需的输出。

grouped_df=df.groupby(['store', 'branch', 'products']).size().unstack('products').replace({np.nan:0})

# output:
products      accessories  bags  clothes  shoes
store branch                                   
1     A               0.0   0.0      1.0    1.0
      C               0.0   0.0      1.0    0.0
2     C               1.0   0.0      1.0    1.0
3     A               0.0   2.0      1.0    0.0
      C               0.0   0.0      1.0    0.0

# desirable output: if (1), (2) and (3) take place somehow...
products      clothes  shoes  accessories  bags
store branch                                   
1     B             0      0            0     0  #group 1 has 1 shoes and 1 clothes for A and C, so 3 in total which transforms each number to 33.3%
      A          33.3   33.3            0     0
      C          33.3    0.0            0     0
2     B             0      0            0     0
      A             0      0            0     0
      C          33.3   33.3         33.3     0
3     B             0      0            0     0  #group 3 has 2 bags and 1 clothes for A and C, so 4 in total which transforms the 2 bags into 50% and so on
      A            25      0            0    50
      C            25      0            0     0
# (3) rearrangement of columns with "clothes" and "shoes" going first
# (3)+(2) branch B appeared and the the order of branches changed to B, A, C
# (1) percentage calculations of the occurrences have been performed over groups that hopefully have made sense with the comments above

我尝试单独处理每个组，但是 i) 它没有考虑替换的 NaN 值，ii) 我应该避免处理每个组，因为之后我需要连接很多组（这个 df 只是一个例子）因为我稍后需要绘制整个组。

grouped_df.loc[[1]].transform(lambda x: x*100/sum(x)).round(0)
>>>
products      accessories  bags  clothes  shoes
store branch                                   
1     A               NaN   NaN     50.0  100.0  #why has it transformed on axis='columns'?
      C               NaN   NaN     50.0    0.0

希望我的问题是有道理的。非常感谢您对我尝试执行的任何见解，非常感谢！

Answer 1

在我发布答案前一天，@Quang Hoang 试图帮助解决这个问题，在我的帮助下，我设法找到了解决方案。

为了解释计算的最后一点，我通过将每个元素除以每个组的计数总和来转换每个元素，以找到每个元素的频率 0th-level-group-wise 而不是 row/column/total-wise .

grouped_df = df.groupby(['store', 'branch', 'products']).size()\
    .unstack('branch')\
        .reindex(['B','C','A'], axis=1, fill_value=0)\
            .stack('branch')\
                .unstack('products')\
                    .replace({np.nan:0})\
                        .transform(
                            lambda x: x*100/df.groupby(['store']).size()
                                   ).round(1)\
                            .reindex(['clothes', 'shoes', 'accessories', 'bags'], axis='columns')

运行上面的代码，产生所需的输出：

products      accessories  bags  clothes  shoes
store branch                                   
1     B               0.0   0.0      0.0    0.0
      C               0.0   0.0     33.3    0.0
      A               0.0   0.0     33.3   33.3
2     B               0.0   0.0      0.0    0.0
      C              33.3   0.0     33.3   33.3
3     B               0.0   0.0      0.0    0.0
      C               0.0   0.0     25.0    0.0
      A               0.0  50.0     25.0    0.0