ENH: general concat with ExtensionArrays through find_common_type (pandas-dev#33607)

Author: jorisvandenbossche

doc/source/whatsnew/v1.1.0.rst

@@ -251,6 +251,9 @@ Backwards incompatible API changes
 - :meth:`DataFrame.at` and :meth:`Series.at` will raise a ``TypeError`` instead of a ``ValueError`` if an incompatible key is passed, and ``KeyError`` if a missing key is passed, matching the behavior of ``.loc[]`` (:issue:`31722`)
 - Passing an integer dtype other than ``int64`` to ``np.array(period_index, dtype=...)`` will now raise ``TypeError`` instead of incorrectly using ``int64`` (:issue:`32255`)
 - Passing an invalid ``fill_value`` to :meth:`Categorical.take` raises a ``ValueError`` instead of ``TypeError`` (:issue:`33660`)
+- Combining a ``Categorical`` with integer categories and which contains missing values
+  with a float dtype column in operations such as :func:`concat` or :meth:`~DataFrame.append`
+  will now result in a float column instead of an object dtyped column (:issue:`33607`)
 
 ``MultiIndex.get_indexer`` interprets `method` argument differently
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

pandas/core/arrays/base.py

@@ -1004,7 +1004,7 @@ def _concat_same_type(
         cls, to_concat: Sequence["ExtensionArray"]
     ) -> "ExtensionArray":
         """
-        Concatenate multiple array.
+        Concatenate multiple array of this dtype.
 
         Parameters
         ----------
@@ -1014,6 +1014,11 @@ def _concat_same_type(
         -------
         ExtensionArray
         """
+        # Implementer note: this method will only be called with a sequence of
+        # ExtensionArrays of this class and with the same dtype as self. This
+        # should allow "easy" concatenation (no upcasting needed), and result
+        # in a new ExtensionArray of the same dtype.
+        # Note: this strict behaviour is only guaranteed starting with pandas 1.1
         raise AbstractMethodError(cls)
 
     # The _can_hold_na attribute is set to True so that pandas internals

pandas/core/arrays/categorical.py

@@ -2296,9 +2296,9 @@ def _can_hold_na(self):
 
     @classmethod
     def _concat_same_type(self, to_concat):
-        from pandas.core.dtypes.concat import concat_categorical
+        from pandas.core.dtypes.concat import union_categoricals
 
-        return concat_categorical(to_concat)
+        return union_categoricals(to_concat)
 
     def isin(self, values):
         """

pandas/core/arrays/integer.py

@@ -1,11 +1,11 @@
 import numbers
-from typing import TYPE_CHECKING, Tuple, Type, Union
+from typing import TYPE_CHECKING, List, Optional, Tuple, Type, Union
 import warnings
 
 import numpy as np
 
 from pandas._libs import lib, missing as libmissing
-from pandas._typing import ArrayLike
+from pandas._typing import ArrayLike, DtypeObj
 from pandas.compat import set_function_name
 from pandas.compat.numpy import function as nv
 from pandas.util._decorators import cache_readonly
@@ -96,6 +96,17 @@ def construct_array_type(cls) -> Type["IntegerArray"]:
         """
         return IntegerArray
 
+    def _get_common_dtype(self, dtypes: List[DtypeObj]) -> Optional[DtypeObj]:
+        # for now only handle other integer types
+        if not all(isinstance(t, _IntegerDtype) for t in dtypes):
+            return None
+        np_dtype = np.find_common_type(
+            [t.numpy_dtype for t in dtypes], []  # type: ignore
+        )
+        if np.issubdtype(np_dtype, np.integer):
+            return _dtypes[str(np_dtype)]
+        return None
+
     def __from_arrow__(
         self, array: Union["pyarrow.Array", "pyarrow.ChunkedArray"]
     ) -> "IntegerArray":

pandas/core/arrays/sparse/array.py

@@ -952,27 +952,7 @@ def copy(self):
 
     @classmethod
     def _concat_same_type(cls, to_concat):
-        fill_values = [x.fill_value for x in to_concat]
-
-        fill_value = fill_values[0]
-
-        # np.nan isn't a singleton, so we may end up with multiple
-        # NaNs here, so we ignore tha all NA case too.
-        if not (len(set(fill_values)) == 1 or isna(fill_values).all()):
-            warnings.warn(
-                "Concatenating sparse arrays with multiple fill "
-                f"values: '{fill_values}'. Picking the first and "
-                "converting the rest.",
-                PerformanceWarning,
-                stacklevel=6,
-            )
-            keep = to_concat[0]
-            to_concat2 = [keep]
-
-            for arr in to_concat[1:]:
-                to_concat2.append(cls(np.asarray(arr), fill_value=fill_value))
-
-            to_concat = to_concat2
+        fill_value = to_concat[0].fill_value
 
         values = []
         length = 0

pandas/core/arrays/sparse/dtype.py

@@ -1,11 +1,13 @@
 """Sparse Dtype"""
 
 import re
-from typing import TYPE_CHECKING, Any, Tuple, Type
+from typing import TYPE_CHECKING, Any, List, Optional, Tuple, Type
+import warnings
 
 import numpy as np
 
-from pandas._typing import Dtype
+from pandas._typing import Dtype, DtypeObj
+from pandas.errors import PerformanceWarning
 
 from pandas.core.dtypes.base import ExtensionDtype
 from pandas.core.dtypes.cast import astype_nansafe
@@ -352,3 +354,23 @@ def _subtype_with_str(self):
         if isinstance(self.fill_value, str):
             return type(self.fill_value)
         return self.subtype
+
+    def _get_common_dtype(self, dtypes: List[DtypeObj]) -> Optional[DtypeObj]:
+
+        fill_values = [x.fill_value for x in dtypes if isinstance(x, SparseDtype)]
+        fill_value = fill_values[0]
+
+        # np.nan isn't a singleton, so we may end up with multiple
+        # NaNs here, so we ignore tha all NA case too.
+        if not (len(set(fill_values)) == 1 or isna(fill_values).all()):
+            warnings.warn(
+                "Concatenating sparse arrays with multiple fill "
+                f"values: '{fill_values}'. Picking the first and "
+                "converting the rest.",
+                PerformanceWarning,
+                stacklevel=6,
+            )
+
+        # TODO also handle non-numpy other dtypes
+        np_dtypes = [x.subtype if isinstance(x, SparseDtype) else x for x in dtypes]
+        return SparseDtype(np.find_common_type(np_dtypes, []), fill_value=fill_value)

pandas/core/dtypes/base.py

@@ -6,6 +6,7 @@
 
 import numpy as np
 
+from pandas._typing import DtypeObj
 from pandas.errors import AbstractMethodError
 
 from pandas.core.dtypes.generic import ABCDataFrame, ABCIndexClass, ABCSeries
@@ -33,11 +34,12 @@ class ExtensionDtype:
     * type
     * name
 
-    The following attributes influence the behavior of the dtype in
+    The following attributes and methods influence the behavior of the dtype in
     pandas operations
 
     * _is_numeric
     * _is_boolean
+    * _get_common_dtype
 
     Optionally one can override construct_array_type for construction
     with the name of this dtype via the Registry. See
@@ -322,3 +324,31 @@ def _is_boolean(self) -> bool:
         bool
         """
         return False
+
+    def _get_common_dtype(self, dtypes: List[DtypeObj]) -> Optional[DtypeObj]:
+        """
+        Return the common dtype, if one exists.
+
+        Used in `find_common_type` implementation. This is for example used
+        to determine the resulting dtype in a concat operation.
+
+        If no common dtype exists, return None (which gives the other dtypes
+        the chance to determine a common dtype). If all dtypes in the list
+        return None, then the common dtype will be "object" dtype (this means
+        it is never needed to return "object" dtype from this method itself).
+
+        Parameters
+        ----------
+        dtypes : list of dtypes
+            The dtypes for which to determine a common dtype. This is a list
+            of np.dtype or ExtensionDtype instances.
+
+        Returns
+        -------
+        Common dtype (np.dtype or ExtensionDtype) or None
+        """
+        if len(set(dtypes)) == 1:
+            # only itself
+            return self
+        else:
+            return None

pandas/core/dtypes/cast.py

@@ -3,7 +3,7 @@
 """
 
 from datetime import date, datetime, timedelta
-from typing import TYPE_CHECKING, Any, Optional, Tuple, Type
+from typing import TYPE_CHECKING, Any, List, Optional, Tuple, Type
 
 import numpy as np
 
@@ -1423,7 +1423,7 @@ def maybe_cast_to_datetime(value, dtype, errors: str = "raise"):
     return value
 
 
-def find_common_type(types):
+def find_common_type(types: List[DtypeObj]) -> DtypeObj:
     """
     Find a common data type among the given dtypes.
 
@@ -1450,8 +1450,16 @@ def find_common_type(types):
     if all(is_dtype_equal(first, t) for t in types[1:]):
         return first
 
+    # get unique types (dict.fromkeys is used as order-preserving set())
+    types = list(dict.fromkeys(types).keys())
+
     if any(isinstance(t, ExtensionDtype) for t in types):
-        return np.object
+        for t in types:
+            if isinstance(t, ExtensionDtype):
+                res = t._get_common_dtype(types)
+                if res is not None:
+                    return res
+        return np.dtype("object")
 
     # take lowest unit
     if all(is_datetime64_dtype(t) for t in types):