如何将pandas中的groupby中的值与先前的值进行比较

Question

我在 pandas 中有以下数据框

  ID     date          no      start      end
  1      01-01-2019    10      101.23     112.23
  2      02-01-2019    10      112.23     120.43
  3      03-01-2019    10      121.23     130.23
  4      04-01-2019    10      130.23     140.43
  5      01-01-2019    11      101        112 
  6      02-01-2019    11      112        120
  7      03-01-2019    11      130        140
  8      04-01-2019    11      140        150.43

我想检查当前行end与下一行start值分组依据no，如果存在差异，则想要设置一个标志并计算差异

以下是我想要的数据框

  ID     date          no      start      end      flag   diff
  1      01-01-2019    10      101.23     112.23   0      0 
  2      02-01-2019    10      112.23     120.43   0      0
  3      03-01-2019    10      121.23     130.23   1      1
  4      04-01-2019    10      130.23     140.43   0      0
  5      01-01-2019    11      101        112      0      0
  6      02-01-2019    11      112        120      0      0
  7      03-01-2019    11      130        140      1      10
  8      04-01-2019    11      140        150.43   0      0

我怎样才能在熊猫中做到这一点？

Answer 1

您可以创建 Series byDataFrameGroupBy.shift并替换第一个NaNs by Series.fillna，比较 bySeries.ne并将掩码转换为整数，对于另一列获得差异：

s = df.groupby('no')['end'].shift().fillna(df['start'])
df['flag'] = df['start'].ne(s).astype(int)
df['diff'] = df['start'] - s
print (df)
   ID        date  no   start     end  flag  diff
0   1  01-01-2019  10  101.23  112.23     0   0.0
1   2  02-01-2019  10  112.23  120.43     0   0.0
2   3  03-01-2019  10  121.23  130.23     1   0.8
3   4  04-01-2019  10  130.23  140.43     0   0.0
4   5  01-01-2019  11  101.00  112.00     0   0.0
5   6  02-01-2019  11  112.00  120.00     0   0.0
6   7  03-01-2019  11  130.00  140.00     1  10.0
7   8  04-01-2019  11  140.00  150.43     0   0.0

细节：

print (s)
0    101.23
1    112.23
2    120.43
3    130.23
4    101.00
5    112.00
6    120.00
7    140.00
Name: end, dtype: float64