CODEWITHSUNDEEP
pythonpandasdataframe
Leon Kyriacou
Leon Kyriacou

Reputation: 402

Optimising Pandas Function That Compares DataFrames

I've got transcation logs which record usage for a kiosk machine and another set of logs for machine online/offline times. The transaction logs contains a datetime field which lets you know when the transaction (or session) occured.

    event_date  raw_data1   session_id  ws_id
0   2017-11-06 12:13:06 {'description': 'Home'} 0604e80d-1ae6-48d0-81bf-32ca1dc58e4c    machine2
1   2017-11-06 12:13:41 {'description': 'AreYouStillThere'} 0604e80d-1ae6-48d0-81bf-32ca1dc58e4c    machine2
2   2017-11-06 12:14:09 {'description': 'AttractiveAnimation'}  0604e80d-1ae6-48d0-81bf-32ca1dc58e4c    machine2
3   2017-11-07 10:06:15 {'description': 'Home'} e2e7565f-60b4-4e7b-a8f0-d0a9c384b283    machine13
4   2017-11-07 10:06:27 {'description': 'AuthenticationPanelAdmin'} e2e7565f-60b4-4e7b-a8f0-d0a9c384b283    machine13

The goal of this function is to see which session_ids conincide with an offline log

    dtrange start   end status  machine_id
0   DateTimeTZRange(datetime.datetime(2017, 11, 17...   2017-11-17 14:46:15 2017-11-17 15:01:15 2   12
1   DateTimeTZRange(datetime.datetime(2017, 11, 17...   2017-11-17 14:47:02 2017-11-17 15:02:02 2   22
2   DateTimeTZRange(datetime.datetime(2017, 11, 17...   2017-11-17 14:47:23 2017-11-17 15:02:23 2   18
3   DateTimeTZRange(datetime.datetime(2017, 11, 17...   2017-11-17 14:48:09 2017-11-17 15:03:09 2   17
4   DateTimeTZRange(datetime.datetime(2017, 11, 17...   2017-11-17 14:49:18 2017-11-17 15:04:18 2   15

ws_id and machine_id are the same, and this makes it a little trickier as the session time and machine_id must match across both dataframes.

This is the code I'm using to return all session_ids that occured when a machine is offline. It filters the offline dataframe with each row from the transaction dataframe and returns a session_id if an offline event coincided with a session time:

def CheckSession(machinename, sessiontime, sessionid):
    if len(offlinedf[(offlinedf.start<sessiontime)
             &(offlinedf.end>sessiontime)
             &(offlinedf.name==machinename)])>0:
        return sessionid

sessions = df.apply(lambda row: CheckSession(row["name"], row["created_at1"], row["session_id"]), axis=1)

This builds the list of sessions, but it is very slow and the dataframes are quite large. I'm still learning how best to work with the pandas library - I was hoping to optimise it using some vectorization but haven't been able to work out how to build it that way.

Upvotes: 1

Views: 73

Answers (1)

Parfait
Parfait

Reputation: 107577

Consider merging df and offlinedf by name and then filter with query according to the logic inside your function. Then convert the filtered dataframe's sessionid column to a list. 

session_df = df.merge(offlinedf, on='name', suffixes=['', '_'])\
               .query('start < created_at1 & end > created_at1') 

sessions = session_df['sessionid'].tolist()

In any data analysis work, blockwise handling of objects are better than iterative row processing.

Upvotes: 1

Related Questions