Timezone object of Timestamp vs DatetimeIndex is different for same time / timezone

[jorisvandenbossche]

Code Sample, a copy-pastable example if possible

When localizing a Timestamp vs timeseries/index, the timezones look similar

In [21]: ts = pd.Timestamp('2012-01-01').tz_localize('US/Eastern')

In [22]: dtidx = pd.to_datetime(['2012-01-01']).tz_localize('US/Eastern')

In [23]: ts
Out[23]: Timestamp('2012-01-01 00:00:00-0500', tz='US/Eastern')

In [24]: dtidx
Out[24]: DatetimeIndex(['2012-01-01 00:00:00-05:00'], dtype='datetime64[ns, US/Eastern]', freq=None)

In [25]: dtidx[0]
Out[25]: Timestamp('2012-01-01 00:00:00-0500', tz='US/Eastern')

but under the hood the tz attributes of the DatetimeIndex is actually different as the tz attribute of the actual Timestamps (even when accessing a single element of the DatetimeIndex as a Timestamp):

In [26]: ts.tz
Out[26]: <DstTzInfo 'US/Eastern' EST-1 day, 19:00:00 STD>

In [27]: dtidx.tz
Out[27]: <DstTzInfo 'US/Eastern' LMT-1 day, 19:04:00 STD>   <-- note the difference here

In [28]: dtidx[0].tz
Out[28]: <DstTzInfo 'US/Eastern' EST-1 day, 19:00:00 STD>

In [29]: ts.tz == dtidx.tz
Out[29]: False

In [30]: ts.tz.zone
Out[30]: 'US/Eastern'

In [31]: dtidx.tz.zone
Out[31]: 'US/Eastern'

[jreback]

of course, these couldn't possibly be the same as in an index you have multiple tzs, while a timestamp has only 1. now sure what you are reporting here.

[jorisvandenbossche]

And I don't understand what you are saying :-)

in an index you have multiple tzs

a DatetimeIndex has a single tz attribute. And even then, why would the tz attribute of a DatetimeIndex be different than the tz attribute of the timestamps inside it ?

[jorisvandenbossche]

Small different example, where I create two equal DatetimeIndex instances in two different ways, and ending up with a different tz object:

# create date_range, then localize
In [51]: dtidx1 = pd.date_range("2012-01-01", periods=3).tz_localize('US/Eastern')

# create same date_range from localized start timestamp
In [52]: dtidx2 = pd.date_range(pd.Timestamp("2012-01-01", tz='US/Eastern'), periods=3)

In [53]: dtidx1.equals(dtidx2)
Out[53]: True

In [54]: dtidx1.tz
Out[54]: <DstTzInfo 'US/Eastern' LMT-1 day, 19:04:00 STD>

In [56]: dtidx2.tz
Out[56]: <DstTzInfo 'US/Eastern' EST-1 day, 19:00:00 STD>

In [57]: dtidx1.tz == dtidx2.tz
Out[57]: False

I am not sure whether is an important issue. I just noticed it, and it seemed odd.

[jreback]

tzobjects themselves don't often compare equal because they are tied to a specific datetime that is the localizer; but the strings do

In [2]: dtidx1.tz
Out[2]: <DstTzInfo 'US/Eastern' LMT-1 day, 19:04:00 STD>

In [3]: dtidx2.tz
Out[3]: <DstTzInfo 'US/Eastern' EST-1 day, 19:00:00 STD>

In [4]: str(dtidx1.tz) == str(dtidx2.tz)
Out[4]: True

In [7]: dtidx1.equals(dtidx2)
Out[7]: True

this is just a fact of life w.r.t. timezone objects. you have entered the rabbit hole.

I think this is a duplicate issue as well, pls have a look.

[mroeschke]

I was confused about this as well at one point (comment), and this SO post echoes @jreback explanation about why the pytz tzobjects reps are different.

While this should probably be explained better in the pytz docs, it might be worth adding a small clarification to the Pandas timeseries docs.

[jorisvandenbossche]

Thanks for that link!
Yeah, I had in the meantime also noticed that the same difference can be seen between an initialized pytz timezone, and the timezone of a datetime localized with it.
So I would say it is more a pytz issue.

Not sure whether this is worth adding a note about it to the pandas docs.

[jreback]

I think this could be worth a small .. note:: section in timeseries/timezones

[AdamShamlian]

I'll take this quickie. Anything to make handling time smoother. I'm thinking something along the lines of mentioning .tz's are not obligated to be the same from Timestamp to "equivalent" DatetimeIndex's and linking both to SO and the other issue?

[jreback]

closing in favor of #18595 which is the underlying reason/cause.