Bug when adding a timedelta to a datetime

[adamdivak]

Hello,

I wanted to add a specified offset (10 seconds) to and existing datetime (an index previously created by pd.to_datetime), so I thought creating a timedelta and adding it would do the trick. As you can see it doesn't yield an error but produces a rather strange results.

@buddha[J:T26]|5> base = pd.to_datetime(datetime.datetime.now())
@buddha[J:T26]|6> offset = pd.to_timedelta(10, unit='s')
@buddha[J:T26]|7> offset
              <7> numpy.timedelta64(10000000000,'ns')
@buddha[J:T26]|8> base
              <8> Timestamp('2014-04-05 13:13:40.374000', tz=None)
@buddha[J:T26]|9> base + offset
              <9> Timestamp('2014-04-05 13:13:41.784065408', tz=None)

If I try to do the same by creating a datetime.timedelta then everything works as expected.

@buddha[J:T26]|10> offset2 = datetime.timedelta(seconds = 10)
@buddha[J:T26]|11> base + offset2
              <11> Timestamp('2014-04-05 13:13:50.374000', tz=None)

This is really confusing and hard to spot. I think that pd.to_datetime and pd.to_timedelta should be compatible, but if they are not, please make the error easier to spot.

Thanks,
Adam

[jreback]

pandas/numpy version?

[jreback]

pandas 0.13.0 (and 0.13.1) / numpy 1.7.1

In [17]: base = pd.to_datetime(datetime.datetime.now())

In [18]: base
Out[18]: Timestamp('2014-04-05 09:55:36.107706', tz=None)

In [19]: offset = pd.to_timedelta(10, unit='s')

In [20]: offset
Out[20]: numpy.timedelta64(10000000000,'ns')

In [21]: offset2 = datetime.timedelta(seconds = 10)

In [22]: offset2
Out[22]: datetime.timedelta(0, 10)

In [23]: base+offset
Out[23]: Timestamp('2014-04-05 09:55:46.107706', tz=None)

In [24]: base+offset2
Out[24]: Timestamp('2014-04-05 09:55:46.107706', tz=None)

[adamdivak]

I'm sorry, forgot to include. Do you think this could be due to the newer numpy version?

commit: None
python: 2.7.6.final.0
python-bits: 32
OS: Windows
OS-release: 7
machine: x86
processor: x86 Family 6 Model 23 Stepping 10, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None

pandas: 0.13.1
Cython: None
numpy: 1.8.0
scipy: 0.13.3
statsmodels: 0.5.0
IPython: 1.2.0
sphinx: 1.2.1
patsy: 0.2.1
scikits.timeseries: None
dateutil: 2.2
pytz: 2013.9
bottleneck: 0.8.0
tables: 3.1.0
numexpr: 2.3
matplotlib: 1.3.1
openpyxl: None
xlrd: 0.9.2
xlwt: 0.7.5
xlsxwriter: None
sqlalchemy: 0.8.4
lxml: 3.3.1
bs4: 4.3.2
html5lib: 0.999
bq: None
apiclient: None

[jreback]

odd....32-bit python messes this up (must be needs to be explicity converted somewhere).

the reason is by default creating an array on 32-bit gives you 32-bit ints, while we use 64-bit ints everywhere

thanks for the report

I guess don't have an explicit test for this

[jreback]

closed via b4f3f0f

I will post updated binaries here soon: http://pandas.pydata.org/pandas-build/dev/

and this will be in 0.14 thanks for the report

[adamdivak]

Thanks for the quick fix.