BooleanArray, StringArray value_counts na_value

[jbrockmendel]

arr = pd.array([True, False, np.nan])

>>> arr
<BooleanArray>
[True, False, NA]
Length: 3, dtype: boolean

>>> arr.value_counts(dropna=False)
True     1
False    1
False    1
dtype: int64

Instead of two False entries, one of them should be pd.NA right? cc @TomAugspurger

arr = pd.array(['foo', 'bar', None])

>>> arr.value_counts(dropna=False)
foo    1
bar    1
NaN    1
dtype: int64

The NaN should be pd.NA right?

Side-note: the defaults for dropna are not consistent across our EAs

[jorisvandenbossche]

That's clearly a bug in the BooleanArray.value_counts implementation. And indeed, we should use pd.NA instead of np.nan in the object index.

[jorisvandenbossche]

@jbrockmendel you have an open PR that relates to EA.value_counts. Does that in some way also fixes this? (otherwise I can have a look at this issue)

[jbrockmendel]

Does that in some way also fixes this? (otherwise I can have a look at this issue)

No, that shouldn't change any behavior.

I think the fix is fairly straightforward for BooleanArray, on L572 np.concatenate([index, np.array([np.nan], dtype=object)]), dtype=object) changing the np.nan to pd.NA should do the trick. Or possibly use self.dtype.na_value like IntegerArray.