Handle FutureWarning from NumPy in Series Construction

[TomAugspurger]

This is with NumPy dev.

In [2]: pd.Series(['0', '1'], dtype=float)
/Users/taugspurger/sandbox/numpy/numpy/core/numeric.py:2468: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison
  return bool(asarray(a1 == a2).all())
Out[2]:
0    0.0
1    1.0
dtype: float64

https://travis-ci.org/pandas-dev/pandas/jobs/427640030#L5344

this goes from core.series._sanitize_array -> _try_cast -> maybe_cast_to_integer_array.

[TomAugspurger]

Also want to look at cmp_method in pandas/core/arrays/datetimelike.py

[jorisvandenbossche]

The occurrences of that last one:

pandas/pandas/core/arrays/datetimelike.py

Lines 59 to 63 in d765dfb

	# numpy will show a DeprecationWarning on invalid elementwise
	# comparisons, this will raise in the future
	with warnings.catch_warnings(record=True):
	with np.errstate(all='ignore'):
	result = op(self.values, np.asarray(other))

pandas/pandas/core/arrays/integer.py

Lines 508 to 512 in 857515f

	# numpy will show a DeprecationWarning on invalid elementwise
	# comparisons, this will raise in the future
	with warnings.catch_warnings(record=True):
	with np.errstate(all='ignore'):
	result = op(self._data, other)

pandas/pandas/core/indexes/base.py

Lines 98 to 102 in d39249a

	# numpy will show a DeprecationWarning on invalid elementwise
	# comparisons, this will raise in the future
	with warnings.catch_warnings(record=True):
	with np.errstate(all='ignore'):
	result = op(self.values, np.asarray(other))

[TomAugspurger]

Looking through this again, I'm concerned I was overly-zealous in my filtering in #22699. In particular, the cmp_method ones are probably bad. We're silencing the following

In [6]: pd.Index([1, 2]) == 'a'
/Users/taugspurger/sandbox/pandas/pandas/core/indexes/base.py:84: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison
  result = op(self.values, np.asarray(other))
Out[6]: False

In [7]: pd.Index([1, 2]) == 1
Out[7]: array([ True, False])

In [8]: pd.Index(['a', 'b']) == 'a'
Out[8]: array([ True, False])

In [9]: pd.Index(['a', 'b']) == 1
Out[9]: array([False, False])

Clearly `Out[6] is the odd one out, so we want NumPy's future behavior. I'm worried now that it'll change without warning, since NumPy was issuing the warnings but we caught them.

How should we handle this? Is it as simple as not filtering those warnings?

[TomAugspurger]

Though I think we're OK pushing this to 0.24.2 or 0.25, since NumPy just had a release.