Always return categorical dtype when get() is called with a mapping

[frankenjoe]

Example:

db = audformat.Database('db')
db.schemes['text'] = audformat.Scheme(dtype='str')
db['misc'] = audformat.MiscTable(
    pd.Index(
      ['a', 'b', 'c'],
      name='speaker',
      dtype='string',
    )
)
db['misc']['text'] = audformat.Column(scheme_id='text')
db['misc']['text'].set(['A', 'B', 'C'])

db.schemes['scheme'] = audformat.Scheme('str', labels='misc')
db['table'] = audformat.MiscTable(audformat.filewise_index(['f1', 'f2', 'f3']))
db['table']['column'] = audformat.Column(scheme_id='scheme')
db['table']['column'].set(['a', 'b', 'c'])

db['table']['column'].get(map='text').dtype

category

and if 'text' contains None, we get:

db['misc']['text'].set(['A', 'B', None])
db['table']['column'].get(map='text').dtype

object

But in both cases we would expect string category.

[hagenw]

It's not straightforward what the expected outcome should be.
string makes sense as this is the dtype of the column in the misc table.

On the other hand labels where handled quite differently before:

db = audformat.Database('db')
db.schemes['scheme'] = audformat.Scheme('str', labels={'a': 'A', 'b': 'B', 'c': 'C'})
db['table'] = audformat.MiscTable(audformat.filewise_index(['f1', 'f2', 'f3']))
db['table']['column'] = audformat.Column(scheme_id='scheme')
db['table']['column'].set(['a', 'b', 'c'])
db['table']['column'].get(map='text').dtype

which returns

CategoricalDtype(categories=['A', 'B', 'C'], ordered=False)

So I guess it some point we need the CategoricalDtype when dealing with labels?

[frankenjoe]

@hagenw is right, the expected behavior would be to have a categorical dtype in both cases. I will rename the issue accordingly.

[hagenw]

A short note, I don't get category as response but the following:

>>> db['table']['column'].get(map='text').dtype
CategoricalDtype(categories=['A', 'B', 'C'], ordered=False)

and the second call I expect to look like this:

>>> db['misc']['text'].set(['A', 'B', None])
>>> db['table']['column'].get(map='text').dtype
CategoricalDtype(categories=['A', 'B'], ordered=False)