DEPR: Categorical with values not present in categories

Author: jbrockmendel

doc/source/whatsnew/v3.0.0.rst

@@ -611,6 +611,7 @@ Other Deprecations
 - Deprecated :meth:`Timestamp.utcfromtimestamp`, use ``Timestamp.fromtimestamp(ts, "UTC")`` instead (:issue:`56680`)
 - Deprecated :meth:`Timestamp.utcnow`, use ``Timestamp.now("UTC")`` instead (:issue:`56680`)
 - Deprecated ``pd.core.internals.api.maybe_infer_ndim`` (:issue:`40226`)
+- Deprecated allowing constructing or casting to :class:`Categorical` with non-NA values that are not present in specified ``dtype.categories`` (:issue:`40996`)
 - Deprecated allowing non-keyword arguments in :meth:`DataFrame.all`, :meth:`DataFrame.min`, :meth:`DataFrame.max`, :meth:`DataFrame.sum`, :meth:`DataFrame.prod`, :meth:`DataFrame.mean`, :meth:`DataFrame.median`, :meth:`DataFrame.sem`, :meth:`DataFrame.var`, :meth:`DataFrame.std`, :meth:`DataFrame.skew`, :meth:`DataFrame.kurt`, :meth:`Series.all`,  :meth:`Series.min`, :meth:`Series.max`, :meth:`Series.sum`, :meth:`Series.prod`, :meth:`Series.mean`, :meth:`Series.median`, :meth:`Series.sem`, :meth:`Series.var`, :meth:`Series.std`, :meth:`Series.skew`, and :meth:`Series.kurt`. (:issue:`57087`)
 - Deprecated allowing non-keyword arguments in :meth:`Series.to_markdown` except ``buf``. (:issue:`57280`)
 - Deprecated allowing non-keyword arguments in :meth:`Series.to_string` except ``buf``. (:issue:`57280`)

pandas/core/arrays/categorical.py

@@ -11,6 +11,7 @@
     cast,
     overload,
 )
+import warnings
 
 import numpy as np
 
@@ -23,6 +24,7 @@
 )
 from pandas._libs.arrays import NDArrayBacked
 from pandas.compat.numpy import function as nv
+from pandas.util._exceptions import find_stack_level
 from pandas.util._validators import validate_bool_kwarg
 
 from pandas.core.dtypes.cast import (
@@ -479,7 +481,11 @@ def __init__(
         elif isinstance(values.dtype, CategoricalDtype):
             old_codes = extract_array(values)._codes
             codes = recode_for_categories(
-                old_codes, values.dtype.categories, dtype.categories, copy=copy
+                old_codes,
+                values.dtype.categories,
+                dtype.categories,
+                copy=copy,
+                warn=True,
             )
 
         else:
@@ -535,7 +541,13 @@ def _from_scalars(cls, scalars, *, dtype: DtypeObj) -> Self:
             # The _from_scalars strictness doesn't make much sense in this case.
             raise NotImplementedError
 
-        res = cls._from_sequence(scalars, dtype=dtype)
+        with warnings.catch_warnings():
+            warnings.filterwarnings(
+                "ignore",
+                "Constructing a Categorical with a dtype and values",
+                FutureWarning,
+            )
+            res = cls._from_sequence(scalars, dtype=dtype)
 
         # if there are any non-category elements in scalars, these will be
         #  converted to NAs in res.
@@ -576,6 +588,15 @@ def astype(self, dtype: AstypeArg, copy: bool = True) -> ArrayLike:
             dtype = self.dtype.update_dtype(dtype)
             self = self.copy() if copy else self
             result = self._set_dtype(dtype, copy=False)
+            wrong = result.isna() & ~self.isna()
+            if wrong.any():
+                warnings.warn(
+                    "Constructing a Categorical with a dtype and values containing "
+                    "non-null entries not in that dtype's categories is deprecated "
+                    "and will raise in a future version.",
+                    FutureWarning,
+                    stacklevel=find_stack_level(),
+                )
 
         elif isinstance(dtype, ExtensionDtype):
             return super().astype(dtype, copy=copy)
@@ -670,14 +691,16 @@ def _from_inferred_categories(
         if known_categories:
             # Recode from observation order to dtype.categories order.
             categories = dtype.categories
-            codes = recode_for_categories(inferred_codes, cats, categories, copy=False)
+            codes = recode_for_categories(
+                inferred_codes, cats, categories, copy=False, warn=True
+            )
         elif not cats.is_monotonic_increasing:
             # Sort categories and recode for unknown categories.
             unsorted = cats.copy()
             categories = cats.sort_values()
 
             codes = recode_for_categories(
-                inferred_codes, unsorted, categories, copy=False
+                inferred_codes, unsorted, categories, copy=False, warn=True
             )
             dtype = CategoricalDtype(categories, ordered=False)
         else:
@@ -1156,7 +1179,7 @@ def set_categories(
             codes = cat._codes
         else:
             codes = recode_for_categories(
-                cat.codes, cat.categories, new_dtype.categories, copy=False
+                cat.codes, cat.categories, new_dtype.categories, copy=False, warn=False
             )
         NDArrayBacked.__init__(cat, codes, new_dtype)
         return cat
@@ -3004,11 +3027,25 @@ def _get_codes_for_values(
     If `values` is known to be a Categorical, use recode_for_categories instead.
     """
     codes = categories.get_indexer_for(values)
+    wrong = (codes == -1) & ~isna(values)
+    if wrong.any():
+        warnings.warn(
+            "Constructing a Categorical with a dtype and values containing "
+            "non-null entries not in that dtype's categories is deprecated "
+            "and will raise in a future version.",
+            FutureWarning,
+            stacklevel=find_stack_level(),
+        )
     return coerce_indexer_dtype(codes, categories)
 
 
 def recode_for_categories(
-    codes: np.ndarray, old_categories, new_categories, *, copy: bool
+    codes: np.ndarray,
+    old_categories,
+    new_categories,
+    *,
+    copy: bool = True,
+    warn: bool = False,
 ) -> np.ndarray:
     """
     Convert a set of codes for to a new set of categories
@@ -3019,6 +3056,8 @@ def recode_for_categories(
     old_categories, new_categories : Index
     copy: bool, default True
         Whether to copy if the codes are unchanged.
+    warn : bool, default False
+        Whether to warn on silent-NA mapping.
 
     Returns
     -------
@@ -3043,9 +3082,18 @@ def recode_for_categories(
             return codes.copy()
         return codes
 
-    indexer = coerce_indexer_dtype(
-        new_categories.get_indexer_for(old_categories), new_categories
-    )
+    codes_in_old_cats = new_categories.get_indexer_for(old_categories)
+    if warn:
+        wrong = codes_in_old_cats == -1
+        if wrong.any():
+            warnings.warn(
+                "Constructing a Categorical with a dtype and values containing "
+                "non-null entries not in that dtype's categories is deprecated "
+                "and will raise in a future version.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
+            )
+    indexer = coerce_indexer_dtype(codes_in_old_cats, new_categories)
     new_codes = take_nd(indexer, codes, fill_value=-1)
     return new_codes
 

pandas/core/groupby/ops.py

@@ -721,7 +721,7 @@ def groups(self) -> dict[Hashable, Index]:
             return self.groupings[0].groups
         result_index, ids = self.result_index_and_ids
         values = result_index._values
-        categories = Categorical(ids, categories=range(len(result_index)))
+        categories = Categorical.from_codes(ids, categories=range(len(result_index)))
         result = {
             # mypy is not aware that group has to be an integer
             values[group]: self.axis.take(axis_ilocs)  # type: ignore[call-overload]

pandas/core/indexes/category.py

@@ -258,6 +258,12 @@ def _is_dtype_compat(self, other: Index) -> Categorical:
         else:
             values = other
 
+            codes = self.categories.get_indexer(values)
+            if ((codes == -1) & ~values.isna()).any():
+                # GH#37667 see test_equals_non_category
+                raise TypeError(
+                    "categories must match existing categories when appending"
+                )
             cat = Categorical(other, dtype=self.dtype)
             other = CategoricalIndex(cat)
             if not other.isin(values).all():

pandas/tests/arrays/categorical/test_api.py

@@ -289,8 +289,16 @@ def test_set_categories(self):
         ],
     )
     def test_set_categories_many(self, values, categories, new_categories, ordered):
-        c = Categorical(values, categories)
-        expected = Categorical(values, new_categories, ordered)
+        msg = "Constructing a Categorical with a dtype and values containing"
+
+        warn1 = FutureWarning if set(values).difference(categories) else None
+        with tm.assert_produces_warning(warn1, match=msg):
+            c = Categorical(values, categories)
+
+        warn2 = FutureWarning if set(values).difference(new_categories) else None
+        with tm.assert_produces_warning(warn2, match=msg):
+            expected = Categorical(values, new_categories, ordered)
+
         result = c.set_categories(new_categories, ordered=ordered)
         tm.assert_categorical_equal(result, expected)
 

pandas/tests/arrays/categorical/test_astype.py

@@ -121,8 +121,11 @@ def test_astype_category(self, dtype_ordered, ordered):
 
         # non-standard categories
         dtype = CategoricalDtype(list("adc"), dtype_ordered)
-        result = cat.astype(dtype)
-        expected = Categorical(data, dtype=dtype)
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            result = cat.astype(dtype)
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            expected = Categorical(data, dtype=dtype)
         tm.assert_categorical_equal(result, expected)
 
         if dtype_ordered is False:

pandas/tests/arrays/categorical/test_constructors.py

@@ -228,14 +228,15 @@ def test_constructor(self):
         # two arrays
         #  - when the first is an integer dtype and the second is not
         #  - when the resulting codes are all -1/NaN
-        with tm.assert_produces_warning(None):
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
             Categorical([0, 1, 2, 0, 1, 2], categories=["a", "b", "c"])
 
-        with tm.assert_produces_warning(None):
+        with tm.assert_produces_warning(FutureWarning, match=msg):
             Categorical([0, 1, 2, 0, 1, 2], categories=[3, 4, 5])
 
         # the next one are from the old docs
-        with tm.assert_produces_warning(None):
+        with tm.assert_produces_warning(FutureWarning, match=msg):
             Categorical([0, 1, 2, 0, 1, 2], [1, 2, 3])
             cat = Categorical([1, 2], categories=[1, 2, 3])
 
@@ -247,12 +248,16 @@ def test_constructor_with_existing_categories(self):
         # GH25318: constructing with pd.Series used to bogusly skip recoding
         # categories
         c0 = Categorical(["a", "b", "c", "a"])
-        c1 = Categorical(["a", "b", "c", "a"], categories=["b", "c"])
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            c1 = Categorical(["a", "b", "c", "a"], categories=["b", "c"])
 
-        c2 = Categorical(c0, categories=c1.categories)
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            c2 = Categorical(c0, categories=c1.categories)
         tm.assert_categorical_equal(c1, c2)
 
-        c3 = Categorical(Series(c0), categories=c1.categories)
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            c3 = Categorical(Series(c0), categories=c1.categories)
         tm.assert_categorical_equal(c1, c3)
 
     def test_constructor_not_sequence(self):
@@ -430,10 +435,13 @@ def test_constructor_dtype_and_others_raises(self):
 
     @pytest.mark.parametrize("categories", [None, ["a", "b"], ["a", "c"]])
     def test_constructor_str_category(self, categories, ordered):
-        result = Categorical(
-            ["a", "b"], categories=categories, ordered=ordered, dtype="category"
-        )
-        expected = Categorical(["a", "b"], categories=categories, ordered=ordered)
+        warn = FutureWarning if categories == ["a", "c"] else None
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(warn, match=msg):
+            result = Categorical(
+                ["a", "b"], categories=categories, ordered=ordered, dtype="category"
+            )
+            expected = Categorical(["a", "b"], categories=categories, ordered=ordered)
         tm.assert_categorical_equal(result, expected)
 
     def test_constructor_str_unknown(self):
@@ -450,10 +458,12 @@ def test_constructor_np_strs(self):
     def test_constructor_from_categorical_with_dtype(self):
         dtype = CategoricalDtype(["a", "b", "c"], ordered=True)
         values = Categorical(["a", "b", "d"])
-        result = Categorical(values, dtype=dtype)
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            result = Categorical(values, dtype=dtype)
         # We use dtype.categories, not values.categories
         expected = Categorical(
-            ["a", "b", "d"], categories=["a", "b", "c"], ordered=True
+            ["a", "b", None], categories=["a", "b", "c"], ordered=True
         )
         tm.assert_categorical_equal(result, expected)
 
@@ -470,16 +480,19 @@ def test_constructor_from_categorical_with_unknown_dtype(self):
     def test_constructor_from_categorical_string(self):
         values = Categorical(["a", "b", "d"])
         # use categories, ordered
-        result = Categorical(
-            values, categories=["a", "b", "c"], ordered=True, dtype="category"
-        )
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            result = Categorical(
+                values, categories=["a", "b", "c"], ordered=True, dtype="category"
+            )
         expected = Categorical(
-            ["a", "b", "d"], categories=["a", "b", "c"], ordered=True
+            ["a", "b", None], categories=["a", "b", "c"], ordered=True
         )
         tm.assert_categorical_equal(result, expected)
 
         # No string
-        result = Categorical(values, categories=["a", "b", "c"], ordered=True)
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            result = Categorical(values, categories=["a", "b", "c"], ordered=True)
         tm.assert_categorical_equal(result, expected)
 
     def test_constructor_with_categorical_categories(self):
@@ -661,17 +674,25 @@ def test_from_inferred_categories_dtype(self):
         cats = ["a", "b", "d"]
         codes = np.array([0, 1, 0, 2], dtype="i8")
         dtype = CategoricalDtype(["c", "b", "a"], ordered=True)
-        result = Categorical._from_inferred_categories(cats, codes, dtype)
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(
+            FutureWarning, match=msg, check_stacklevel=False
+        ):
+            result = Categorical._from_inferred_categories(cats, codes, dtype)
         expected = Categorical(
-            ["a", "b", "a", "d"], categories=["c", "b", "a"], ordered=True
+            ["a", "b", "a", None], categories=["c", "b", "a"], ordered=True
         )
         tm.assert_categorical_equal(result, expected)
 
     def test_from_inferred_categories_coerces(self):
         cats = ["1", "2", "bad"]
         codes = np.array([0, 0, 1, 2], dtype="i8")
         dtype = CategoricalDtype([1, 2])
-        result = Categorical._from_inferred_categories(cats, codes, dtype)
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(
+            FutureWarning, match=msg, check_stacklevel=False
+        ):
+            result = Categorical._from_inferred_categories(cats, codes, dtype)
         expected = Categorical([1, 1, 2, np.nan])
         tm.assert_categorical_equal(result, expected)
 
@@ -722,7 +743,9 @@ def test_interval(self):
 
         # extra
         values = pd.interval_range(8, 11, periods=3)
-        cat = Categorical(values, categories=idx)
+        msg = "Constructing a Categorical with a dtype and values containing"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
+            cat = Categorical(values, categories=idx)
         expected_codes = np.array([8, 9, -1], dtype="int8")
         tm.assert_numpy_array_equal(cat.codes, expected_codes)
         tm.assert_index_equal(cat.categories, idx)