time delta index

[hayd]

You can't seem to create a time delta index, even if you force object dtype:

In [4]: td = pd.Series([np.timedelta64(10000), pd.NaT], dtype='timedelta64[ns]')

In [5]: pd.Index(td)
Out[5]: Int64Index([10000, -9223372036854775808], dtype='int64')

In [6]: pd.Index(td, dtype= 'timedelta64[ns]')
Out[6]: Int64Index([10000, -9223372036854775808], dtype='int64')

related to #7423

Perhaps there is already an issue for this?

[cpcloud]

There's an issue for implementing time delta indexes but not for making object indexes with them. Can you create one with Python timedeltas?

[jreback]

a time delta index impl is not that hard
it's very much like DatetimeIndex

[cpcloud]

Yep I just need to carve out some time for it among the other million issues :) I think split and replace Apis are higher priority

[shoyer]

Here is the existing issue for TimedeltaIndex: #3009

You actually can make an object index with a list of np.timedelta64 objects, but not from an ndarray:

>>> pd.Index(list(td))
Index([10000 nanoseconds, NaT], dtype='object')

[hayd]

thanks, closing as dupe