我正在尝试在数据框中添加一行。条件是当用户再次回到该应用程序(300秒后)时,我需要添加一行。下面是我的代码。它可以正常工作,但需要花费大量执行时间,因为实际数据帧有1000万行。
cursor.fetchall()
输入:
for i in range(1,len(df)):
if df['user_id'][i]==df['user_id'][i-1] and (df['start_time'][i]-df['start_time'][i-1]).seconds>300:
df.loc[len(df)]=[df['user_id'][i],df['start_time'][i],'psuedo_App_start_2']
输出应如下所示:
user_id start_time event
100 03/04/19 6:11 psuedo_App_start
100 03/04/19 6:11 notification_receive
100 03/04/19 8:56 notification_dismiss
10 03/04/19 22:05 psuedo_App_start
10 03/04/19 22:05 subcategory_click
10 03/04/19 22:06 subcategory_click
从输出中可以看到,user_id = 100添加了一行,因为他回到8.56,即300秒后返回。
答案 0 :(得分:2)
首先按2个条件进行过滤-比较user_id
和每个组的DataFrameGroupBy.shift
个值,还比较每个组的差异和DataFrameGroupBy.diff
,然后重新分配evet
列和{{3 }},最后DataFrame.assign
在一起,并按concat
排序:
#MM/DD/YY HH:MM
#df['start_time'] = pd.to_datetime(df['start_time'])
#DD/MM/YY HH:MM
#df['start_time'] = pd.to_datetime(df['start_time'], dayfirst=True)
m1 = df['user_id'].eq(df.groupby('user_id')['user_id'].shift())
m2 = df.groupby('user_id')['start_time'].diff().dt.total_seconds() > 300
df1 = df[m1 & m2].assign(event='psuedo_App_start_2')
df1 = (pd.concat([df, df1], ignore_index=True)
.sort_values(['user_id','start_time'], ascending=[False, True]))
print (df1)
user_id start_time event
0 100 2019-03-04 06:11:00 psuedo_App_start
1 100 2019-03-04 06:11:00 notification_receive
2 100 2019-03-04 08:56:00 notification_dismiss
6 100 2019-03-04 08:56:00 psuedo_App_start_2
3 10 2019-03-04 22:05:00 psuedo_App_start
4 10 2019-03-04 22:05:00 subcategory_click
5 10 2019-03-04 22:06:00 subcategory_click
答案 1 :(得分:0)
通常,在这种情况下,您需要将显式循环转换为矢量化操作。尝试这样的事情:
i = (df.user_id.values[1:] == df.user_id.values[:-1]) & ((df.start_time.values[1:] - df.start_time.values[:-1])/np.timedelta64(1, 's') > 300)
newRows = tt[np.append(False, i)].copy()
newRows.event = 'psuedo_App_start_2'
df.append(newRows)