pd.DataFrame.from_dict fails when index is numpy.datetime

[JohnNapier]

Suppose that we want to create a dataframe from a dictionary.

import pandas as pd,numpy as np
#1) Form dataframe
ix=pd.date_range('1/1/2015',periods=5,freq='D')
df0=pd.DataFrame(np.random.normal(size=(len(ix),5)),index=ix)
#2) Split dataframe into 2 dictionaries, based on sign
d={'+':{},'-':{}}
for loc in df0.index:
    d['+'][loc]=df0.loc[loc][df0.loc[loc]>=0]
    d['-'][loc]=df0.loc[loc][df0.loc[loc]<0]
#3) Convert dictionary to dataframe
print pd.DataFrame.from_dict(d['+'],orient='index')

This returns a dataframe of positive values, with NaN in place of the negative values. Now, it appears the exact same operation fails if we use "df.index.values" instead of "df.index".

d={'+':{},'-':{}}
for loc in df0.index.values: # NOTE .values!
    d['+'][loc]=df0.loc[loc][df0.loc[loc]>=0]
    d['-'][loc]=df0.loc[loc][df0.loc[loc]<0]
print pd.DataFrame.from_dict(d['+'],orient='index')

Now the outcome is a dataframe with all None. I think the problem is some inconsistency between pandas' datetime and numpy's datetime formats.

Pandas version 0.16.1, Numpy version 1.9.2. Thanks.

[jreback]

This is a bug here. Pull-requests are welcome. In any event, you should prob just do this.

In [7]: pd.concat({ '+' : df0[df0>=0], '-' : df0[df0<0] })
Out[7]: 
                     0         1         2         3         4
+ 2015-01-01       NaN       NaN  0.592870  0.007121       NaN
  2015-01-02  0.104371  0.618107  1.347137       NaN       NaN
  2015-01-03       NaN  1.477950       NaN       NaN       NaN
  2015-01-04       NaN  1.588150  0.172968       NaN       NaN
  2015-01-05  0.631164  1.921521  0.952758       NaN       NaN
- 2015-01-01 -1.248147 -1.011006       NaN       NaN -0.230722
  2015-01-02       NaN       NaN       NaN -1.097534 -0.768210
  2015-01-03 -1.541076       NaN -0.160582 -0.277739 -1.420439
  2015-01-04 -1.584375       NaN       NaN -0.304879 -0.263297
  2015-01-05       NaN       NaN       NaN -0.450119 -0.230233

[jreback]

closed by #10269