[ Convert timedelta64[ns] column to seconds in Python Pandas DataFrame ]

A pandas DataFrame column duration contains timedelta64[ns] as shown. How can you convert them to seconds?

0   00:20:32
1   00:23:10
2   00:24:55
3   00:13:17
4   00:18:52
Name: duration, dtype: timedelta64[ns]

I tried the following

print df[:5]['duration'] / np.timedelta64(1, 's')

but got the error

Traceback (most recent call last):
  File "test.py", line 16, in <module>
    print df[0:5]['duration'] / np.timedelta64(1, 's')
  File "C:\Python27\lib\site-packages\pandas\core\series.py", line 130, in wrapper
    "addition and subtraction, but the operator [%s] was passed" % name)
TypeError: can only operate on a timedeltas for addition and subtraction, but the operator [__div__] was passed

Also tried

print df[:5]['duration'].astype('timedelta64[s]')

but received the error

Traceback (most recent call last):
  File "test.py", line 17, in <module>
    print df[:5]['duration'].astype('timedelta64[s]')
  File "C:\Python27\lib\site-packages\pandas\core\series.py", line 934, in astype
    values = com._astype_nansafe(self.values, dtype)
  File "C:\Python27\lib\site-packages\pandas\core\common.py", line 1653, in _astype_nansafe
    raise TypeError("cannot astype a timedelta from [%s] to [%s]" % (arr.dtype,dtype))
TypeError: cannot astype a timedelta from [timedelta64[ns]] to [timedelta64[s]]

Answer 1

This works properly in the current version of Pandas (version 0.14):

In [132]: df[:5]['duration'] / np.timedelta64(1, 's')
Out[132]: 
0    1232
1    1390
2    1495
3     797
4    1132
Name: duration, dtype: float64

Here is a workaround for older versions of Pandas/NumPy:

In [131]: df[:5]['duration'].values.view('<i8')/10**9
Out[131]: array([1232, 1390, 1495,  797, 1132], dtype=int64)

timedelta64 and datetime64 data are stored internally as 8-byte ints (dtype '<i8'). So the above views the timedelta64s as 8-byte ints and then does integer division.

Note that you need NumPy version 1.7 or newer to work with datetime64/timedelta64s.