groupby.first() casts datetime64[ns] series to 'object' with 'long" elements

[gerigk]

import numpy as np
import pandas as pd
df = pd.DataFrame([(1, 1351036800000000000), (2, 1351036800000000000)])
df[1] = df[1].view('M8[ns]')
print df[1].dtype
print df.groupby(0).first()[1].dtype
print type(df.groupby(0).first()[1][1])

datetime64[ns]
object
<type 'long'>

[wesm]

There's a couple issues here. looking into it

[wesm]

Dear lord don't look at what I had to do to fix this in a simple/dirty way

[gerigk]

@wesm I don't know whether this reopens the issue:

I received the following error after updating pandas:

ind_part = df[time_dim].groupby(level=group_dims).first()
  File "/usr/local/lib/python2.7/dist-packages/pandas-0.9.2.dev_4519310-py2.7-linux-x86_64.egg/pandas/core/groupby.py", line 38, in f
    result = result.convert_objects()
AttributeError: 'Series' object has no attribute 'convert_objects'

I fixed it by using

DataFrame(df[time_dim]).groupby(level=group_dims).first()

which kind of lets me think that your fix only works for DataFrames but not for Series.

[wesm]

I'll have a look

[changhiskhan]

Please don't look at this one either :)

[changhiskhan]

closed via a28a5cc