API: deprecate get_offset, Offset for time rules like '30s'

[jseabold]

From stackoverflow: http://stackoverflow.com/questions/17579932/statsmodel-arma-function-is-incompatiable-with-pandas/17581170

I'm looking only quickly at this, but I'm wondering if this should be recognized and return an offset.

from pandas.tseries.frequencies import get_offset
get_offset('30s')

or if in statsmodels we should accept custom offsets in place of frequency strings.

[hayd]

Perhaps poor implementation (seems very hacky), but perhaps something like:

In [11]: s = '30S'

In [12]: m = re.match('(?P<number>\d+)(?P<offset>.*)', s)

In [13]: int(m.groupdict()['number']) * get_offset(m.groupdict()['offset'])
Out[13]: <30 Seconds>

should be done on the pandas side.

[jseabold]

That's what I had in mind as long as it's valid for any integer + existing offset. I'll go ahead and add a workaround on our side like this.

[jreback]

All kinds of goodies in there.....is this in the docs?

In [1]: from pandas.tseries.frequencies import to_offset

In [2]: to_offset('30s')
Out[2]: <30 Seconds>

[jreback]

def get_offset(name):
    """
    Return DateOffset object associated with rule name

    Examples
    --------
    get_offset('EOM') --> BMonthEnd(1)
    """
def to_offset(freqstr):
    """
    Return DateOffset object from string representation

    Examples
    --------
    >>> to_offset('5Min')
    Minute(5)
    """

[jseabold]

Ah, great. Any need to keep these functions separate? get_ seems like a subset of to_ to me (without looking at the source).

[jreback]

that sounds right.....so changing this issue to deprectate get_offset..thxs