如何根据条件行筛选行？

Question

我有一个按时间排序的数据帧：

  trID   event  time frameID    
  tID1    d      t1   1
  tID1    s      t2   1
  tID1    d      t3   1     
  tID1    d      t4   2      
  tID1    s      t5   2
  tID1    d      t6   1      
  tID1    s      t7   1      

我想得到max（time）-min（time）之差的和 每次frameID更改时。对于上面显示的数据帧，预期结果是： （t3-t1）+（t7-t6）

有多个trID，每个trID我都需要它

Answer 1

用途：

print (df)
   trID event       time  frameID
0  tID1     d 2015-01-01        1
1  tID1     s 2015-01-02        1
2  tID1     d 2015-01-03        1
3  tID1     d 2015-01-04        2
4  tID1     s 2015-01-05        2
5  tID1     d 2015-01-06        1
6  tID1     s 2015-01-07        1

#create helper column for distinguish frameID with duplicates for unique groups
df['g'] = df['frameID'].ne(df['frameID'].shift()).cumsum()

#aggregate by 3 columns ang get difference
a = df.groupby(['trID','g','frameID'])['time'].agg(lambda x: x.max() - x.min())
print (a)
trID  g  frameID
tID1  1  1         2 days
      2  2         1 days
      3  1         1 days
Name: time, dtype: timedelta64[ns]

#sum by first and third level (trID, frameID) - get days
df1 = a.sum(level=[0,2]).dt.days.reset_index(name='SUM')
#similar get seconds if necessary
#df1 = a.sum(level=[0,2]).dt.total_seconds().reset_index(name='SUM')
print (df1)
   trID  frameID  SUM
0  tID1        1    3
1  tID1        2    1

编辑：

print (df)
   trID event        time  frameID
0  tID1     d  2015-01-01        2
1  tID1     s  2015-01-02        2
2  tID1     d  2015-01-03        1
3  tID1     d  2015-01-04        1
4  tID2     s  2015-01-05        1
5  tID2     d  2015-01-06        1
6  tID1     s  2015-01-07        1
7  tID1     s  2015-01-08        1

df['time'] = pd.to_datetime(df['time'])

#create 2 columns for groups
df[['g1','g2']] = df[['trID','frameID']].ne(df[['trID','frameID']].shift()).cumsum()

#add both new columns
a = df.groupby(['trID','g1','g2','frameID'])['time'].agg(lambda x: x.max() - x.min())
print (a)
trID  g1  g2  frameID
tID1  1   1   2         1 days
          2   1         1 days
      3   2   1         1 days
tID2  2   2   1         1 days
Name: time, dtype: timedelta64[ns]

#sum by first and fourth level (trID, frameID)
df1 = a.sum(level=[0,3]).dt.days.reset_index(name='SUM')
print (df1)
   trID  frameID  SUM
0  tID1        2    1
1  tID1        1    2
2  tID2        1    1