一种基于组的日间隔为列分配值的优雅方法

时间:2019-08-27 09:19:59

标签: python python-3.x pandas

我有一个如下所示的数据框

df1 = pd.DataFrame({'subject_id' :[1,1,1,1,1,1,1,2,2,2,2],'day':[3,7,9,10,11,19,20,7,13,18,22] , 'fake_flag' :['fake VAC','','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC']})

看起来如下图

enter image description here

我想根据以下规则在actual_flag列中填充值

a)fake_flag的值为fake_vac,并且不能为空

b)仅在出现fake_vac的第一天和14 days interval之后的记录中填充值。

这是我尝试过的

t = df1[df1['fake_flag'] == 'fake VAC']
sub_list = t['subject_id'].unique().tolist()
   for sub in sub_list:
     day_list = t['day'][t['subject_id']==sub].tolist()
     min_value = min(day_list)
     index = t[t['day']==min_value].index
     df1.loc[index, 'actual_flag'] = 'act_vac'
     i_14day = min_value + 14
     day_values = [i for i in day_list if i >= i_14day]
     print("day greater than 14 are ", day_values)
     if len(day_values) > 0:
         for val in day_values:
            index = t[t['day']==val].index
            df1.loc[index, 'actual_flag'] = 'act_vac'

如您所见,这很长,对于百万条记录的数据集,我无法做到这一点。任何有效而优雅的方法都是有帮助的

期望我的输出如下图所示

enter image description here

在这种情况下,对于subject_id = 1,day 3是第一次出现fake vacday 19(从3开始间隔为19天> 14天)和day 20 (20 gt> 14天间隔,从3开始)在14天间隔之后。任何优雅高效的解决方案都是有帮助的

用于测试的示例数据

df1 = pd.DataFrame({'subject_id' :[1,1,1,1,1,1,1,1,2,2,2,2],'day':[2,3,7,9,10,11,19,20,7,13,18,22] , 'fake_flag' :['','fake VAC','','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC','fake VAC']})

**更新了屏幕截图**

enter image description here

1 个答案:

答案 0 :(得分:1)

一种方法是从每个组中的所有日期中减去第一天,检查大于14的日期并将其设置为"act_vac"以及初始日期:

import numpy as np
# Returns a boolean with True if a given day - first day > 14
ix = df1.fake_flag.ne('').groupby(df1.subject_id).transform('idxmax')
c1 = df1.day.sub(df1.values[ix, 1]).gt(14)
# True if the id is different to previous row
c2 = df1.subject_id.ne(df1.subject_id.shift())
# logical OR of the above conditions
df1['actual_flag'] = np.where(c1 | c2, 'act_vac', '')

     subject_id  day fake_flag actual_flag
0            1    3  fake VAC     act_vac
1            1    7                      
2            1    9  fake VAC            
3            1   10  fake VAC            
4            1   11  fake VAC            
5            1   19  fake VAC     act_vac
6            1   20  fake VAC     act_vac
7            2    7  fake VAC     act_vac
8            2   13  fake VAC            
9            2   18  fake VAC            
10           2   22  fake VAC     act_vac

详细信息

df1.assign(c1=c1, c2=c2, actual_flag= np.where(c1 | c2, 'act_vac', ''))

     subject_id  day fake_flag actual_flag     c1     c2
0            1    3  fake VAC     act_vac  False   True
1            1    7                        False  False
2            1    9  fake VAC              False  False
3            1   10  fake VAC              False  False
4            1   11  fake VAC              False  False
5            1   19  fake VAC     act_vac   True  False
6            1   20  fake VAC     act_vac   True  False
7            2    7  fake VAC     act_vac  False   True
8            2   13  fake VAC              False  False
9            2   18  fake VAC              False  False
10           2   22  fake VAC     act_vac   True  False