BUG: frame creation with multi-index and tuples in dict

[ruidc]

In [1]: import pandas;import numpy;import datetime;
In [2]: v = datetime.date.today()
In [3]: pandas.DataFrame({v : pandas.Series(range(3),index=range(3))}, columns=[v])
Out[3]:
   2015-08-20
0           0
1           1
2           2
In [4]: v = v, v
In [5]: pandas.DataFrame({v : pandas.Series(range(3),index=range(3))}, columns=[v])
Out[5]:
  (2015-08-20, 2015-08-20)
0                      NaN
1                      NaN
2                      NaN

[kawochen]

See #7888

use datetime.datetime.combine(datetime.date.today(), datetime.time()) instead of datetime.date.today()

[jreback]

yep, so the issue is with the datetime.date, which is mostly unsupported.

In [9]: v = Timestamp.today().normalize()

In [10]: v
Out[10]: Timestamp('2015-09-08 00:00:00')

In [11]: pandas.DataFrame({v : pandas.Series(range(3),index=range(3))}, columns=[v])
Out[11]: 
   2015-09-08
0           0
1           1
2           2

In [12]: v = v, v                                                                   

In [13]: pandas.DataFrame({v : pandas.Series(range(3),index=range(3))}, columns=[v])
Out[13]: 
   (2015-09-08 00:00:00, 2015-09-08 00:00:00)
0                                           0
1                                           1
2                                           2

[n8henrie]

I assume I am running into this same issue -- multiindex with pd.Timestamp.date() in a tuple results in all NaN, pd.Timestamp works fine.

Examples below don't have anything else in the tuple to simplify demonstration of the issue, but my use case obviously has other stuff there.

In [3]: pd.DataFrame({'foo': {(pd.Timestamp(2017, 10, 5).date(),): 21.0}})
Out[3]:
            foo
2017-10-05  NaN

In [4]: pd.DataFrame({'foo': {(pd.Timestamp(2017, 10, 5),): 21.0}})
Out[4]:
            foo
2017-10-05  21.0

I can go into detail on my use case if it were helpful, but it's basically looking at employee productivity per shift, where each day (part 0 of the index) has several shifts (part 1 of the index).

For now, using Timestamp.normalize() seems to be a workaround, even though putting all the events at 00:00 seems a little strange.

[mroeschke]

The issue works on master. Could use a test.

In [34]: In [4]: v = v, v
    ...: In [5]: pandas.DataFrame({v : pandas.Series(range(3),index=range(3))}, columns=[v])
Out[34]:
   (2019-10-13, 2019-10-13)
0                         0
1                         1
2                         2

In [35]: pd.__version__
Out[35]: '0.26.0.dev0+555.gf7d162b18'