Convert Pandas TimeDelta to integer

Question

Suppose I have a Pandas Series that contains TimeDelta data. In fact it has been generated by taking the difference of a DateTimeIndex with a shifted version of itself, hence giving the delta between consecutive timestamps.

It looks something like

timestamp
2015-02-01 00:00:04   00:00:04
2015-02-01 00:00:08   00:00:04
2015-02-01 00:00:12   00:00:04
....
Name: timestamp, dtype: timedelta64[ns]

The values are obviously numpy.timedelta64 but I need to get them into seconds. There have been similar questions asked relating to this but no answers I have seen yet that deals with Pandas 0.16.1.

What I've tried is:

ts.apply(lambda x: x.seconds)

Which gives an error of

AttributeError: 'numpy.timedelta64' object has no attribute 'seconds'

Then tried

numpy.int64(ts)

But that gives me an array. Now I know I can convert that back into a Series but is there not another way to do this in one Pandas call or mapping function?

Answer 1

The following worked for me:

In [24]:

t="""index,timestamp
2015-02-01 00:00:04,00:00:04
2015-02-01 00:00:08,00:00:04
2015-02-01 00:00:12,00:00:04"""
s = pd.read_csv(io.StringIO(t),parse_dates=[0,1], squeeze=True, index_col=[0])
In [26]:

s.dt.second
Out[26]:
index
2015-02-01 00:00:04    4
2015-02-01 00:00:08    4
2015-02-01 00:00:12    4
dtype: int64

datetime dtype values have a dt accessor where you can access the seconds attribute.