ERR: Index.shift() gives confusing error message when no DatetimeIndex

[jorisvandenbossche]

In [1]: idx = pd.Index(range(5))

In [2]: idx
Out[2]: Int64Index([0, 1, 2, 3, 4], dtype='int64')

In [3]: idx.shift(1)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-3-d8362983dbf7> in <module>()
----> 1 idx.shift(1)

c:\users\vdbosscj\scipy\pandas-joris\pandas\core\index.pyc in shift(self, period
s, freq)
   1096             return self
   1097
-> 1098         offset = periods * freq
   1099         return Index([idx + offset for idx in self], name=self.name)
   1100

TypeError: unsupported operand type(s) for *: 'int' and 'NoneType'

Should shift be allowed at all when it is no DatetimeIndex?

[jreback]

this is a bug as the freq=None. I am not sure of the usecase of this for non-datetimelike (though its seems to 'work')

e.g.

idx.shift(periods=1,freq=1)

[jreback]

In [6]: i = Index(range(5))

In [7]: i
Out[7]: Int64Index([0, 1, 2, 3, 4], dtype='int64')

In [8]: i.shift(periods=1,freq=1)
Out[8]: Int64Index([1, 2, 3, 4, 5], dtype='int64')

In [9]: i.shift(periods=1,freq=2)
Out[9]: Int64Index([2, 3, 4, 5, 6], dtype='int64')

In [10]: i.shift(periods=1,freq=3)
Out[10]: Int64Index([3, 4, 5, 6, 7], dtype='int64')

In [11]: i.shift(periods=2,freq=3)
Out[11]: Int64Index([6, 7, 8, 9, 10], dtype='int64')

In [12]: i = date_range('20130101',periods=5)

In [13]: i
Out[13]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-01, ..., 2013-01-05]
Length: 5, Freq: D, Timezone: None

In [15]: i.shift(1)
Out[15]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-02, ..., 2013-01-06]
Length: 5, Freq: D, Timezone: None

In [16]: i.shift(2)
Out[16]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-03, ..., 2013-01-07]
Length: 5, Freq: D, Timezone: None

In [17]: i.shift(2,freq='2D')
Out[17]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-05, ..., 2013-01-09]
Length: 5, Freq: None, Timezone: None

[jorisvandenbossche]

yes, you can indeed 'misuse' the shift function on a normal Int64Index, eg also:

In [10]: idx.shift(3, freq=1.5)
Out[10]: Float64Index([4.5, 5.5, 6.5, 7.5, 8.5], dtype='float64')

but I don't really know what the use of this could be (just idx + 1 instead)? And if we even should allow it, as I think a user expects something else from a shift function (how it works on a series, shifting the values)

So maybe raising on a non-Datetime/PeriodIndex?

BTW, what is best way to really shift an index?

[jreback]

closed by #11211