Pandas convert_dtypes need support convert string with nan to int .

Is your feature request related to a problem?

convert_dtypes need support convert string with nan to int

Describe the solution you'd like

Now, pandas doesn't consider string with nan case as numerical variable .

pd.Series(['1', '2', '3', np.nan]).convert_dtypes().dtype

got

string[python]

Describe alternatives you've considered

pd.Series(['1', '2', '3', np.nan]).convert_dtypes().dtype

should like pd.Series(['1', '2', '3', np.nan]).astype('Int16').dtype

Int16Dtype()

Comment From: isaacangyu

take

Comment From: isaacangyu

This my first comment on an issue, but after some researching on convert_dtypes(), I got the impression that convert_dtypes() was only meant to convert object data types(https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.convert_dtypes.html). Since your list ['1', '2', '3', np.nan] contained strings, not objects, convert_dtypes() could not perform the desired function. However, you may have been making a suggestion for a new feature, in which case I support. It might help if you clarify. I hope my comment was helpful!

Comment From: eromoe

Yeah, this is a feature request.

Comment From: isaacangyu

Thanks for clarifying! I support this feature request and will unassign myself.

Comment From: mroeschke

IMO I think pd.Series(['1', '2', '3', np.nan]).convert_dtypes() returning string[python] is appropriate here. The object type contains strings and np.nan which is the missing value indicator` so string is the appropriate conversion since pandas has a string dtype now. I would be -1 on inferring integer here.

Comment From: phofl

agreed, -1 on converting to int as well. Those are strings, so should keep as is