Performance drop when using timezone-aware DateTimeIndex

[adrien-pain-01]

It seems that pandas.DataFrame operations on Index with timezone-aware dates is order of magnitude slower than on regular datetimes.

for a 500k datetimes created with pandas.date_range, and using DataFrame.shift() to compute deltas between dates, timings goes from 17ms for standard datetimes to 16seconds for timezone-aware datetimes.

I don't understand why it is so slow with timezones objects.

I already posted a complete message related to this behavior on stackoverflow yesterday :
http://stackoverflow.com/questions/30385481/performance-of-timezone-aware-pandas-datetimeindex

I'm using latest pandas 0.16.1 from Anaconda, and latest numpy 1.9.2

[jreback]

see #8260 as this is a known issue.

datetimes with tz's are represented as object dtype, rather than datetime64[ns for datetimes. The fix is to implement #8260 which not difficult is a bit in depth as it requires learning a bit about the internals.

[adrien-pain-01]

@jreback:
thanks for your explanation and quick reply.
will wait then for the next major release to see the fix :-)

and really good work on pandas, awesome library !

[jreback]

closed by #10477