Losing Nanosecond precision upon conversion to DatetimeIndex

[lterry-t]

All,

I am experiencing a problem when converting from a TimeSeries to a DatetimeIndex. In the code below, I create a TimeSeries of nanosecond precision TimeStamps. I then convert it to a DatetimeIndex and lose the nanosecond precision (but maintain the microseconds). This appears to be a bug as the resulting elements of DatetimeIndex are of time TimeStamp and can handle nanosecond precision. Please see the snippet below.

Any help/resolution would be greatly appreciated.

Thanks,
Joe

In [233]: t1 = pandas.Timestamp((1352934390*1000000000)+1000000+1000+1)

In [234]: t2 = pandas.Timestamp((1352934390*1000000000)+1000000+1000+5)

In [235]: timeseries = pandas.TimeSeries([t1,t2])

In [236]: timeseries[0]
Out[236]: <Timestamp: 2012-11-14 23:06:30.001001001>

In [237]: datetimeindex = pandas.DatetimeIndex(timeseries)

In [238]: datetimeindex[0]
Out[238]: <Timestamp: 2012-11-14 23:06:30.001001>

In [236]: timeseries[0]
Out[236]: <Timestamp: 2012-11-14 23:06:30.001001001>

[kajdogg]

Joe, I think your point is obscured by the markup behavior on < and >. I see the same puzzling behavior :

t1 = pandas.Timestamp((1352934390_1000000000)+1000000+1000+1)
t2 = pandas.Timestamp((1352934390_1000000000)+1000000+1000+5)
timeseries = pandas.TimeSeries([t1,t2])
timeseries[0]
<Timestamp: 2012-11-14 23:06:30.001001001>

datetimeindex = pandas.DatetimeIndex(timeseries)
datetimeindex[0]
<Timestamp: 2012-11-14 23:06:30.001001>
timeseries[0]
<Timestamp: 2012-11-14 23:06:30.001001001>

the last two timestamps are not the same

[wesm]

Looks like a bug. Will have to wait for 0.9.2 as I just cut 0.9.1 release-- will fix in the dev version as soon as we can

[lterry-t]

Kajdogg, thanks for fixing. Wes, thanks for the quick response! I'll keep an eye out for it.

[angryoxen]

Not sure how big your series is, but getting the raw epoch nanoseconds out of the Timestamps works, if you have time for the extraction:

import pandas
t1 = pandas.Timestamp((1352934390_1000000000)+1000000+1000+1)
t2 = pandas.Timestamp((1352934390_1000000000)+1000000+1000+5)
timeseries = pandas.TimeSeries([t1,t2])
datetimeindex = pandas.DatetimeIndex([t.value for t in timeseries])
datetimeindex[0]

=> Timestamp: 2012-11-14 23:06:30.001001001

[changhiskhan]

This should be fixed on master now. Do you want to give it a shot?

[lterry-t]

Works great now. Thanks for the quick fix!!

[changhiskhan]

sweet. thanks for the report