我有以下数据帧:
id_x id_y
department date
0 09/2017 1 NaN
1 01/2018 149 NaN
01/2019 112 4.0
02/2018 103 1.0
02/2019 78 NaN
... ... ...
799 09/2017 57 2.0
10/2017 64 3.0
11/2017 80 NaN
12/2017 79 2.0
这是从数据库数据构造的数据帧的结果,在该数据中,运行了一系列计数并按部门和日期进行分组。
我需要按部门和日期汇总的这些数据,但是,我希望日期跨越顶部,然后是 id 计数。
我想要的输出大致如下:
9/2017 10/2017
id_x id_y id_x id_y
department
0 1 NaN NaN NaN
1 NaN NaN NaN NaN
... ... ... ... ...
799 57 2.0 64 3.0
我尝试删除索引、重新编制索引、融化数据帧和透视数据帧。 我可以按"id_x"和"id_y"后跟日期来排序数据帧,但是,这不是一个优雅的解决方案,因为它为每个 id 重复了 36 个日期。
我一直在引用以下文档: https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.pivot.html
并测试了以下解决方案的变体(除其他外):
new_df.melt(new_df, col_level=0, id_vars=['department'], value_vars=['id_x','id_y'])
new_df.reset_index().pivot_table(index="department", columns="date") #I've also tried "date" as values and in brackets outside the parenthesis
重新创建了您的数据,但我认为这可以满足您的需求? 如果日期字段在 df 中实际上是日期时间,则排序将按升序日期顺序显示数据帧。
df=pd.DataFrame({'department':[0,1,1,1,1,799,799,799,799],'date':['09/2017','01/2018','01/2019','02/2018','02/2019','09/2017','10/2017','11/2017','12/2017'],'id_x':[1,149,112,103,78,57,64,80,79],'id_y':[np.NaN,np.NaN,4.0,1.0,np.NaN,2.0,3.0,np.NaN,2.0]})
df=df.set_index('department')
df2=df.pivot(columns='date',values=['id_x','id_y'])
df3=df2.swaplevel(axis=1)
df3.sort_index(axis=1, level=0, inplace=True)
输出:
date 01/2018 01/2019 02/2018 ... 10/2017 11/2017 12/2017
id_x id_y id_x id_y id_x ... id_y id_x id_y id_x id_y
department ...
0 NaN NaN NaN NaN NaN ... NaN NaN NaN NaN NaN
1 149.0 NaN 112.0 4.0 103.0 ... NaN NaN NaN NaN NaN
799 NaN NaN NaN NaN NaN ... 3.0 80.0 NaN 79.0 2.0