BUG: to_numeric doesn't work uint64 numbers

[verhalenn]

uint64 isn't very well supported right now but something to consider.

In [10]: pd.to_numeric(pd.Series([0, 9223372036854775808]), downcast = 'unsigned')
---------------------------------------------------------------------------
OverflowError                             Traceback (most recent call last)
<ipython-input-10-6e8272095758> in <module>()
----> 1 pd.to_numeric(pd.Series([0, 9223372036854775808]), downcast = 'unsigned')

/home/verhalenn/Documents/Open-Source/pandas/pandas/tools/util.py in to_numeric(arg, errors, downcast)
    193             coerce_numeric = False if errors in ('ignore', 'raise') else True
    194             values = lib.maybe_convert_numeric(values, set(),
--> 195                                                coerce_numeric=coerce_numeric)
    196 
    197     except Exception:

/home/verhalenn/Documents/Open-Source/pandas/pandas/src/inference.pyx in pandas.lib.maybe_convert_numeric (pandas/lib.c:53043)()
    667             seen_float = True
    668         elif util.is_integer_object(val):
--> 669             floats[i] = ints[i] = val
    670             seen_int = True
    671         elif util.is_bool_object(val):

OverflowError: Python int too large to convert to C long

[gfyoung]

@jreback : The bug can be traced to this line here, where we try to store uint64 in an int64 array. Do we need to create another array for integers beyond np.iinfo(np.int64).max?

[jreback]

yep, we don't have lots of tests / support for uint64's atm. so pull requests to fix welcome.

[gfyoung]

@jreback : What would be the best way to patch do you think? The problematic case comes when we have negative integers and positive integers exceed np.iinfo(np.int64).max:

>>> import numpy as np
>>> np.array([-1, np.iinfo(np.int64).max + 1])  # Even numpy has issues.
array([ -1.00000000e+00,   9.22337204e+18])

[mroeschke]

Looks to work on master. Could use a test.

In [73]: In [10]: pd.to_numeric(pd.Series([0, 9223372036854775808]), downcast = 'unsigned')
    ...:
Out[73]:
0                      0
1    9223372036854775808
dtype: uint64

In [74]: pd.__version__
Out[74]: '0.26.0.dev0+684.g953757a3e'