hash_pandas_object fails on empty dataframe with index=False

[TomAugspurger]

In [8]: df = pd.DataFrame(index=pd.Index([1, 2, 3]))

In [9]: pd.util.hash_pandas_object(df, index=False)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-9-23d57ac9e66b> in <module>
----> 1 pd.util.hash_pandas_object(df, index=False)

~/sandbox/pandas/pandas/core/util/hashing.py in hash_pandas_object(obj, index, encoding, hash_key, categorize)
    112         h = _combine_hash_arrays(hashes, num_items)
    113
--> 114         h = Series(h, index=obj.index, dtype='uint64', copy=False)
    115     else:
    116         raise TypeError("Unexpected type for hashing %s" % type(obj))

~/sandbox/pandas/pandas/core/series.py in __init__(self, data, index, dtype, name, copy, fastpath)
    244                             'Length of passed values is {val}, '
    245                             'index implies {ind}'
--> 246                             .format(val=len(data), ind=len(index)))
    247                 except TypeError:
    248                     pass

ValueError: Length of passed values is 0, index implies 3

I'm not sure what to do here. I think that the output should have the same shape / index. Passing index=False means don't include the index in the hashed value. But the output should retain the same index.

[TomAugspurger]

Actually, I think the only sensible result is an empty dataframe with the same index right? In other words, the exact same thing as the input.

[nathalier]

Hi,
I'd like to start contributing to pandas (again, after maternity break) so I'm looking for not complicated issues. Possibly this could fit:-)
I've found that hash_pandas_object() actually returns series and not dataframe.
So is that really OK that dataframe would be return in that special case?

[TomAugspurger]

I've found that hash_pandas_object() actually returns series and not dataframe.

@nathalier you're correct. An empty series would be better to return.

[nathalier]

@TomAugspurger missing data are automatically substituted with NaN in series. Could there be further inconsistencies if we return series with NaN as hash values?

[TomAugspurger]

Sorry, I don’t follow the question. This issue is specifically about *empty* DataFrames, not a DataFrame with all or some missing values.

…

On Feb 3, 2019, at 22:04, nathalier ***@***.***> wrote: @TomAugspurger missing data are automatically substituted with NaN in series. Could there be further inconsistencies if we return series with NaN as hash values? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[nathalier]

I'm talking about Series which should be returned.
Possibly I'm wrong but as I see empty Series with indices cannot be created. I.e.,
In[3]: pd.Series(index=pd.Index([1, 2, 3]))
Out[3]: 1 NaN 2 NaN 3 NaN

Or empty series without indices is what's expected?

[TomAugspurger]

Oh sorry, you mean the output. I missed the "return" in your comment.

Yes that is a difficulty, since we need to return a uint64-dtype series. Off top of my head, I'm not sure what's best here...