Average time for datetime list

Question

Looking for fastest solution of time averaging problem.

I've got a list of datetime objects. Need to find average value of time (excluding year, month, day). Here is what I got so far:

import datetime as dtm
def avg_time(times):
    avg = 0
    for elem in times:
        avg += elem.second + 60*elem.minute + 3600*elem.hour
    avg /= len(times)
    rez = str(avg/3600) + ' ' + str((avg%3600)/60) + ' ' + str(avg%60)
    return dtm.datetime.strptime(rez, "%H %M %S")

Answer 1

Here's a better way to approach this problem

Generate a sample of datetimes

In [28]: i = date_range('20130101',periods=20000000,freq='s')

In [29]: i
Out[29]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-01 00:00:00, ..., 2013-08-20 11:33:19]
Length: 20000000, Freq: S, Timezone: None

avg 20m times

In [30]: %timeit pd.to_timedelta(int((i.hour*3600+i.minute*60+i.second).mean()),unit='s')
1 loops, best of 3: 2.87 s per loop

The result as a timedelta (note that this requires numpy 1.7 and pandas 0.13 for the to_timedelta part, coming very soon)

In [31]: pd.to_timedelta(int((i.hour*3600+i.minute*60+i.second).mean()),unit='s')
Out[31]: 
0   11:59:12
dtype: timedelta64[ns]

In seconds (this will work for pandas 0.12, numpy >= 1.6).

In [32]: int((i.hour*3600+i.minute*60+i.second).mean())
Out[32]: 43152