order not preserved when constructing DataFrame from a list of OrderedDicts

[ilrico]

issue is explained here:
http://stackoverflow.com/questions/37484738/pandas-dataframe-construction-from-a-list-of-ordereddict-preserving-columns-ord?noredirect=1#comment62465925_37484738

workaround with constructor option "columns" works but not elegant.

[jreback]

pls post a copy-pastable reproducible example at the top of the PR.

yes this could be an enhancement. pull-requests are welcome.

[gfyoung]

Here's one:

>>> from pandas.compat import OrderedDict
>>> from pandas import DataFrame
>>>
>>> data = OrderedDict()
>>> data['c'] = 2
>>> data['b'] = 1
>>>
>>> data
OrderedDict([('c', 2), ('b', 1)])
>>>
>>> DataFrame([data])
   b  c
0  1  2

[gfyoung]

I should point out though that if you pass in the OrderedDict with arrays as the values, the order is in fact preserved!

>>> from pandas.compat import OrderedDict
>>> from pandas import DataFrame
>>>
>>> data = OrderedDict()
>>> data['c'] = [2]
>>> data['b'] = [1]
>>>
>>> data
OrderedDict([('c', [2]), ('b', [1])])
>>>
>>> DataFrame(data)
   c  b
0  2  1

[gfyoung]

The "bug" traces back to the initialization of arrays from the list of dictionaries. The initialization sorts the keys at the end, so we just need a boolean to tell it not to sort when we have OrderedDict. Can take this one home. A welcomed change of scenery from the land of read_csv! 😄