BUG: Fix maybe_convert_numeric for unhashable objects

[RogerThomas]

• closes to_numeric(errors='coerce') broken when using unhashable objects #13324
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

[jreback]

add the issue number here as a comment

[jreback]

test usage of pd.to_numeric as well (tests in pandas/tools

[codecov-io]

Current coverage is 84.22%

Merging #13326 into master will increase coverage by <.01%

@@             master     #13326   diff @@
==========================================
  Files           138        138          
  Lines         50684      50681     -3   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
- Hits          42684      42682     -2   
+ Misses         8000       7999     -1   
  Partials          0          0          

Powered by Codecov. Last updated by c0850ea...80b8515

[jreback]

Bug in pd.to_numeric when errors='coerce' and input contains non-hashable objects....

[RogerThomas]

@jreback i've made the changes you suggested

[jreback]

use tm.assert_numpy_array_equal (we don't use th numpy testing functions at all; in fact there is a check again them; though this one slipped thru)

[jreback]

looks good. ping when green.

[jreback]

@RogerThomas thanks!

I was just experimenting with something and:

so maybe should change this to if getattr(val, '__hash__', None) is not None

though can't seem to get this to trigger

In [1]: 'a'.__hash__
Out[1]: <method-wrapper '__hash__' of str object at 0x1002e1508>

In [2]: None.__hash__
Out[2]: <method-wrapper '__hash__' of NoneType object at 0x100180c40>

In [3]: 1.__hash__
  File "<ipython-input-3-e97b483155ff>", line 1
    1.__hash__
             ^
SyntaxError: invalid syntax

In [5]: [].__hash__ is None
Out[5]: True

[jreback]

if you want to see if you can generate a test that makes this fail (not sure). If so , can you do a followup?

[RogerThomas]

Hey @jreback, I think this is just due to python's syntax parser.

In [1]: a = 1

In [2]: a.__hash__
Out[2]: <method-wrapper '__hash__' of int object at 0xa63b00>

I don't think this could ever fail

[jreback]

oh ok that makes sense thanks