NaT in MultiIndex throws ValueError when reset_index() is called.

[jstritar]

Please see this test case:

x = pp.DataFrame({
    'other': ['a', np.nan],
    'date': ['2015-03-22', np.nan],
    'amount': [2, 3]
  })
x.date = pp.to_datetime(x.date)

index	amount	date	other
0	2	2015-03-22	a
1	3	NaT	NaN

The reset_index here then fails with the error below:

x.set_index(['other', 'date']).reset_index()

ValueError: Converting an integer to a NumPy datetime requires a specified unit

pp.version.version
'0.16.1'
pp.np.version.version
'1.7.1'

[jreback]

hmm, thought we had an issue for this, guess not. This core.common._maybe_upcast_putmask is kind of wonky (was needed when Series was a ndarray sub-class), but I think can be replaced but a block operation (I think).

[jstritar]

Do you recommend any work around?