我有一个带有对话和时间戳的数据框,如下所示:
timestamp userID textBlob new_id
2018-10-05 23:07:02 01 a large text blob...
2018-10-05 23:07:13 01 a large text blob...
2018-10-05 23:07:23 01 a large text blob...
2018-10-05 23:07:36 01 a large text blob...
2018-10-05 23:08:02 01 a large text blob...
2018-10-05 23:09:16 01 a large text blob...
2018-10-05 23:09:21 01 a large text blob...
2018-10-05 23:09:39 01 a large text blob...
2018-10-05 23:09:47 01 a large text blob...
2018-10-05 23:10:01 01 a large text blob...
2018-10-05 23:10:11 01 a large text blob...
2018-10-05 23:10:23 01 restart
2018-10-05 23:10:59 01 a large text blob...
2018-10-05 23:11:03 01 a large text blob...
2018-10-08 23:11:32 02 a large text blob...
2018-10-08 23:12:58 02 a large text blob...
2018-10-08 23:13:16 02 a large text blob...
2018-10-08 23:14:04 02 a large text blob...
2018-10-08 03:38:36 02 a large text blob...
2018-10-08 03:38:42 02 a large text blob...
2018-10-08 03:38:52 02 a large text blob...
2018-10-08 03:38:57 02 a large text blob...
2018-10-08 03:39:10 02 a large text blob...
2018-10-08 03:39:27 02 Restart
2018-10-08 03:40:47 02 a large text blob...
2018-10-08 03:40:54 02 a large text blob...
2018-10-08 03:41:02 02 a large text blob...
2018-10-08 03:41:12 02 a large text blob...
2018-10-08 03:41:32 02 a large text blob...
2018-10-08 03:41:39 02 a large text blob...
2018-10-08 03:42:20 02 a large text blob...
2018-10-08 03:44:58 02 a large text blob...
2018-10-08 03:45:54 02 a large text blob...
2018-10-08 03:46:06 02 a large text blob...
2018-10-08 05:06:42 03 a large text blob...
2018-10-08 05:06:53 03 a large text blob...
2018-10-08 05:08:49 03 a large text blob...
2018-10-08 05:08:58 03 a large text blob...
2018-10-08 05:58:18 04 a large text blob...
2018-10-08 05:58:26 04 a large text blob...
2018-10-08 05:58:37 04 a large text blob...
2018-10-08 05:58:58 04 a large text blob...
2018-10-08 06:00:31 04 a large text blob...
2018-10-08 06:01:00 04 a large text blob...
2018-10-08 06:01:14 04 a large text blob...
2018-10-08 06:02:03 04 a large text blob...
2018-10-08 06:02:03 04 a large text blob...
2018-10-08 06:06:03 04 a large text blob...
2018-10-08 06:10:00 04 a large text blob...
2018-10-08 09:07:03 04 a large text blob...
2018-10-08 09:09:03 04 a large text blob...
2018-10-09 10:01:00 04 a large text blob...
2018-10-09 10:02:00 04 a large text blob...
2018-10-09 10:03:00 04 a large text blob...
2018-10-09 10:09:00 04 a large text blob...
2018-10-09 10:09:00 05 a large text blob...
此刻,我想用ID识别数据框内的对话。问题在于用户可以有多个对话(即userID
可以有多个textBlob
关联)。因此,我想添加一个new_id
以便能够识别上述数据框内的对话。
为此,我想基于三个条件创建一个new_id
列:
预期输出如下(*)
:
timestamp userID textBlob new_id
2018-10-05 23:07:02 01 a large text blob... 001
2018-10-05 23:07:13 01 a large text blob... 001
2018-10-05 23:07:23 01 a large text blob... 001
2018-10-05 23:07:36 01 a large text blob... 001
2018-10-05 23:08:02 01 a large text blob... 001
2018-10-05 23:09:16 01 a large text blob... 001
2018-10-05 23:09:21 01 a large text blob... 001
2018-10-05 23:09:39 01 a large text blob... 001
2018-10-05 23:09:47 01 a large text blob... 001
2018-10-05 23:10:01 01 a large text blob... 001
2018-10-05 23:10:11 01 a large text blob... 001
2018-10-05 23:10:23 01 restart 001 ---- (The word restart appeared so a new id is created ↓)
2018-10-05 23:10:59 01 a large text blob... 002
2018-10-05 23:11:03 01 a large text blob... 002
2018-10-08 23:11:32 02 a large text blob... 002
2018-10-08 23:12:58 02 a large text blob... 002
2018-10-08 23:13:16 02 a large text blob... 002
2018-10-08 23:14:04 02 a large text blob... 002 --- (The conversation ends because the 10 minutes time threshold was exceeded)
2018-10-08 03:38:36 02 a large text blob... 003
2018-10-08 03:38:42 02 a large text blob... 003
2018-10-08 03:38:52 02 a large text blob... 003
2018-10-08 03:38:57 02 a large text blob... 003
2018-10-08 03:39:10 02 a large text blob... 003
2018-10-08 03:39:27 02 Restart 003 ---- (The word restart appeared so a new id is created ↓)
2018-10-08 03:40:47 02 a large text blob... 004
2018-10-08 03:40:54 02 a large text blob... 004
2018-10-08 03:41:02 02 a large text blob... 004
2018-10-08 03:41:12 02 a large text blob... 004
2018-10-08 03:41:32 02 a large text blob... 004
2018-10-08 03:41:39 02 a large text blob... 004
2018-10-08 03:42:20 02 a large text blob... 004
2018-10-08 03:44:58 02 a large text blob... 004
2018-10-08 03:45:54 02 a large text blob... 004
2018-10-08 03:46:06 02 a large text blob... 004 ---- (The 10 minutes threshold is exceeded a new id is assigned ↓)
2018-10-08 05:06:42 03 a large text blob... 005
2018-10-08 05:06:53 03 a large text blob... 005
2018-10-08 05:08:49 03 a large text blob... 005
2018-10-08 05:08:58 03 a large text blob... 005 ---- (no more conversations from user id 03, thus the a new id is assigned)
2018-10-08 05:58:18 04 a large text blob... 006
2018-10-08 05:58:26 04 a large text blob... 006
2018-10-08 05:58:37 04 a large text blob... 006
2018-10-08 05:58:58 04 a large text blob... 006
2018-10-08 06:00:31 04 a large text blob... 006
2018-10-08 06:01:00 04 a large text blob... 006
2018-10-08 06:01:14 04 a large text blob... 006
2018-10-08 06:02:03 04 a large text blob... 006 ---- (The 10 minutes threshold is exceeded a new id is assigned ↓)
2018-10-08 06:02:03 04 a large text blob... 007
2018-10-08 06:06:03 04 a large text blob... 007
2018-10-08 06:10:00 04 a large text blob... 007
2018-10-08 09:07:03 04 a large text blob... 007
2018-10-08 09:09:03 04 a large text blob... 007 ---- (The 10 minutes threshold is exceeded a new id is assigned ↓)
2018-10-09 10:01:00 04 a large text blob... 008
2018-10-09 10:02:00 04 a large text blob... 008
2018-10-09 10:03:00 04 a large text blob... 008
2018-10-09 10:09:00 04 a large text blob... 008 ---- (no more conversations from user id 04, thus the a new id is assigned)
2018-10-09 10:09:00 05 a large text blob... 010
到目前为止,我试图:
searchfor = ['restart','Restart']
df['keyword_id'] = df['textBlob'].str.contains('|'.join(searchfor))
和
dif = df['timestamp'] - df['timestamp'].shift()
periods = dif > pd.Timedelta('10 min')
times = periods.cumsum().apply(lambda x: x+1)
df['time_id'] = times
但是,我还需要考虑userID,最后我得到了几列。有什么办法可以满足这三个条件并获得预期的输出(*)
?
答案 0 :(得分:1)
您已到达那里。综上所述,为每个条件构建一个布尔掩码,然后将掩码转换为int并取其累积和:
mask1 = df.timestamp.diff() > pd.Timedelta(10, 'm')
mask2 = df['userID'].diff() != 0
mask3 = df['textBlob'].shift().str.lower() == 'restart'
df['new_id'] = (mask1 | mask2 | mask3).astype(int).cumsum()
# Result:
print(df.to_string(index=False))
timestamp userID textBlob new_id
2018-10-05 23:07:02 1 a_large_text_blob... 1
2018-10-05 23:07:13 1 a_large_text_blob... 1
2018-10-05 23:07:23 1 a_large_text_blob... 1
2018-10-05 23:07:36 1 a_large_text_blob... 1
2018-10-05 23:08:02 1 a_large_text_blob... 1
2018-10-05 23:09:16 1 a_large_text_blob... 1
2018-10-05 23:09:21 1 a_large_text_blob... 1
2018-10-05 23:09:39 1 a_large_text_blob... 1
2018-10-05 23:09:47 1 a_large_text_blob... 1
2018-10-05 23:10:01 1 a_large_text_blob... 1
2018-10-05 23:10:11 1 a_large_text_blob... 1
2018-10-05 23:10:23 1 restart 1
2018-10-05 23:10:59 1 a_large_text_blob... 2
2018-10-05 23:11:03 1 a_large_text_blob... 2
2018-10-08 03:11:32 2 a_large_text_blob... 3
2018-10-08 03:12:58 2 a_large_text_blob... 3
2018-10-08 03:13:16 2 a_large_text_blob... 3
2018-10-08 03:14:04 2 a_large_text_blob... 3
2018-10-08 03:38:36 2 a_large_text_blob... 4
2018-10-08 03:38:42 2 a_large_text_blob... 4
2018-10-08 03:38:52 2 a_large_text_blob... 4
2018-10-08 03:38:57 2 a_large_text_blob... 4
2018-10-08 03:39:10 2 a_large_text_blob... 4
2018-10-08 03:39:27 2 Restart 4
2018-10-08 03:40:47 2 a_large_text_blob... 5
2018-10-08 03:40:54 2 a_large_text_blob... 5
2018-10-08 03:41:02 2 a_large_text_blob... 5
2018-10-08 03:41:12 2 a_large_text_blob... 5
2018-10-08 03:41:32 2 a_large_text_blob... 5
2018-10-08 03:41:39 2 a_large_text_blob... 5
2018-10-08 03:42:20 2 a_large_text_blob... 5
2018-10-08 03:44:58 2 a_large_text_blob... 5
2018-10-08 03:45:54 2 a_large_text_blob... 5
2018-10-08 03:46:06 2 a_large_text_blob... 5
2018-10-08 05:06:42 3 a_large_text_blob... 6
2018-10-08 05:06:53 3 a_large_text_blob... 6
2018-10-08 05:08:49 3 a_large_text_blob... 6
2018-10-08 05:08:58 3 a_large_text_blob... 6
2018-10-08 05:58:18 4 a_large_text_blob... 7
2018-10-08 05:58:26 4 a_large_text_blob... 7
2018-10-08 05:58:37 4 a_large_text_blob... 7
2018-10-08 05:58:58 4 a_large_text_blob... 7
2018-10-08 06:00:31 4 a_large_text_blob... 7
2018-10-08 06:01:00 4 a_large_text_blob... 7
2018-10-08 06:01:14 4 a_large_text_blob... 7
2018-10-08 06:02:03 4 a_large_text_blob... 7
2018-10-08 06:02:03 4 a_large_text_blob... 7
2018-10-08 06:06:03 4 a_large_text_blob... 7
2018-10-08 06:10:00 4 a_large_text_blob... 7
2018-10-08 09:07:03 4 a_large_text_blob... 8
2018-10-08 09:09:03 4 a_large_text_blob... 8
2018-10-09 10:01:00 4 a_large_text_blob... 9
2018-10-09 10:02:00 4 a_large_text_blob... 9
2018-10-09 10:03:00 4 a_large_text_blob... 9
2018-10-09 10:09:00 4 a_large_text_blob... 9
2018-10-09 10:09:00 5 a_large_text_blob... 10
答案 1 :(得分:-1)
好吧,我认为10分钟的时间应该从会话开始算起,而不是从下面的直接消息算起,在这种情况下,您需要遍历以下行:
df['timestamp'] = pd.to_datetime(df['timestamp'])
restart = df.textBlob.str.contains('|'.join(['restart','Restart']))
user_change = df.userID == df.userID.shift().fillna(method='bfill')
df['new_id'] = (restart | ~user_change).cumsum()
current_id = 0
new_id_prev = 0
start_time = df.timestamp.iloc[0]
for i, new_id, timestamp in zip(range(len(df)), df.new_id, df.timestamp):
timedelta = timestamp - start_time
if new_id != new_id_prev or timedelta > pd.Timedelta(10,unit='m'):
current_id += 1
start_time = timestamp
new_id_prev = new_id
df.new_id.iloc[i] = current_id