TimedeltaIndex constructor gives wild output

[ahjulstad]

import pandas as pd
a = pd.TimedeltaIndex(range(3),unit='S')
print (a[1])

gives output

11574 days 01:46:40

I would expect the same output as from

import pandas as pd
import numpy as np
npa = np.array(range(3),dtype='timedelta64[s]')
a = pd.TimedeltaIndex(npa)
print(a[1])

giving output

 0 days 00:00:01

This is pandas 0.15.1 as installed by conda update pandas from a fresh python 3.4 install of anaconda.

INSTALLED VERSIONS
------------------
commit: None
python: 3.4.1.final.0
python-bits: 64
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 58 Stepping 9, GenuineIntel
byteorder: little
LC_ALL: None
LANG: nb_NO

pandas: 0.15.1
nose: 1.3.4
Cython: 0.21
numpy: 1.9.1
scipy: 0.14.0
statsmodels: 0.5.0
IPython: 2.2.0
sphinx: 1.2.3
patsy: 0.3.0
dateutil: 2.1
pytz: 2014.9
bottleneck: None
tables: 3.1.1
numexpr: 2.3.1
matplotlib: 1.4.0
openpyxl: 1.8.5
xlrd: 0.9.3
xlwt: None
xlsxwriter: 0.5.7
lxml: 3.4.0
bs4: 4.3.2
html5lib: None
httplib2: None
apiclient: None
rpy2: None
sqlalchemy: 0.9.7
pymysql: None
psycopg2: None

[jreback]

this is a dupe of #8886

you can do this

In [1]: pd.to_timedelta(range(3),unit='s')
Out[1]: 
<class 'pandas.tseries.tdi.TimedeltaIndex'>
['00:00:00', ..., '00:00:02']
Length: 3, Freq: None

the unit paramater should be accepted directly by TimedeltaIndex with an implementation that passes to to_timedelta.

care to do a pull-request?

[ahjulstad]

From what I can understand, the unit parameter is already accepted by TimedeltaIndex, and sent to to_timedelta. I checked the returned value, it is 1e18 ns, indicating that the scaling factor (between s and ns) is applied twice. The problem seems to be here somewhere, it calls to_timedelta twice.

       if unit is not None:
            data = to_timedelta(data, unit=unit, box=False)
[snipped away]
        # convert if not already
        if getattr(data,'dtype',None) != _TD_DTYPE:
            data = to_timedelta(data,unit=unit,box=False)
            print('second time',data)

(This is code from 4f8af44#diff-c03d3accabc2e2f441d87b961f4425d7R172 )

I am not sure this is a duplicate, but then again I don't understand all the points in the original issue report.

[ahjulstad]

The following changes fixes my problem, it appears, but I do not understand fully why.

Link to branch in my github tree:
https://github.com/ahjulstad/pandas/tree/fix-9011

(I have added a test, as well.)

[jreback]

I think you can ust not pass unit to this line: https://github.com/pydata/pandas/blob/master/pandas/tseries/tdi.py#L184 (its already converted if it has a passed unit).

go ahead and take your test and make the change do a pull-request. If you can do this shortly, can include it in 0.15.2

[jreback]

ideally also add some more tests cases enumerated in #8886