unexpected behavior of pd.core.algorithms._ensure_data()

[realead]

Examples

Example 1:

import pandas.core.algorithms as algos
comps = ['ss', 42]
values = ['42']
algos.isin(comps, values)

results in array([False, True], dtype=bool)

Example 2:

import pandas.core.algorithms as algos
comps = ['ss', np.nan]
values = [np.nan]
algos.isin(comps, values)

results in array([False, False], dtype=bool)

Expected Output

The results should be [False, False] for example 1 and [False, True] for example 2.

Problem description

The problem can be tracked down to this line: https://github.com/pandas-dev/pandas/blob/master/pandas/core/algorithms.py#L137

    # we have failed, return object
    values = np.asarray(values)
   return ensure_object(values), 'object', 'object'

np.asarray(values) tries to do smart things (like converting 42 to a string), it should be probably avoided by passing dtype=np.object to it.

Output of pd.show_versions()

pandas 0.22.0

[realead]

This is an aspect of #22119 and probably related to #22148.

[gfyoung]

@realead : Agreed. Feel free to investigate the (unrelated) portion of this bug / issue.