How do I round datetime column to nearest quarter hour

I have loaded a data file into a Python pandas dataframe. I has a datetime column of the format 2015-07-18 13:53:33.280.

What I need to do is create a new column that rounds this out to its nearest quarter hour. So, the date above will be rounded to 2015-07-18 13:45:00.000.

How do I do this in pandas? I tried using the solution from here, but get an 'Series' object has no attribute 'year' error.

Solution 1:

You can use round(freq). There is also a shortcut column.dt for datetime functions access (as @laurens-koppenol suggests).

Here's one-liner:

df['old column'].dt.round('15min')  

String aliases for valid frequencies can be found here. Full working example:

In [1]: import pandas as pd    
In [2]: df = pd.DataFrame([pd.Timestamp('2015-07-18 13:53:33.280'),
                           pd.Timestamp('2015-07-18 13:33:33.330')],
                         columns=['old column'])

In [3]: df['new column']=df['old column'].dt.round('15min')  
In [4]: df
Out[4]: 
               old column          new column
0 2015-07-18 13:53:33.280 2015-07-18 14:00:00
1 2015-07-18 13:33:33.330 2015-07-18 13:30:00

Solution 2:

Assuming that your series is made up of datetime objects, You need to use Series.apply . Example -

import datetime
df['<column>'] = df['<column>'].apply(lambda dt: datetime.datetime(dt.year, dt.month, dt.day, dt.hour,15*(dt.minute // 15)))

The above example to always round to the previous quarter hour (behavior similar to floor function).

EDIT

To round to the correct quarter hour (as in , if its 7 mins 30 seconds past previous quarter, to show the next quarter) . We can use the below example -

import datetime
df['<column>'] = df['<column>'].apply(lambda dt: datetime.datetime(dt.year, dt.month, dt.day, dt.hour,15*round((float(dt.minute) + float(dt.second)/60) / 15)))

The above would only take the latest seconds into consideration , if you want the millisecond/microsecond into consideration , you can add that to the above equation as - (float(dt.minute) + float(dt.second)/60 + float(dt.microsecond)/60000000)