Add fix to raise error when category value 'x' is not predefined but is assigned through df.loc[..]=x

pandas/core/generic.py

@@ -65,6 +65,7 @@
     ensure_str,
     is_bool,
     is_bool_dtype,
+    is_categorical_dtype,
     is_datetime64_any_dtype,
     is_datetime64tz_dtype,
     is_dict_like,
@@ -5228,6 +5229,19 @@ def __setattr__(self, name: str, value) -> None:
         After regular attribute access, try setting the name
         This allows simpler access to columns for interactive use.
         """
+        # Fix for handling assignment to a Catagorical dtype
+        # with a category which is not predefined.
+        if isinstance(value, BlockManager) and is_categorical_dtype(self):
+            if len(value) > 0:
+                new_value = value.as_array()[-1]
+            else:
+                new_value = None
+            if not pd.isna(new_value) and new_value not in self.dtype.categories.values:
+                raise ValueError(
+                    "Cannot setitem on a Categorical with a new "
+                    "category, set the categories first"
+                )
+
         # first try regular attribute access via __getattribute__, so that
         # e.g. ``obj.x`` and ``obj.x = 4`` will always reference/modify
         # the same attribute.

pandas/tests/arrays/categorical/test_indexing.py

@@ -234,6 +234,22 @@ def test_where_ordered_differs_rasies(self):
         with pytest.raises(ValueError, match="without identical categories"):
             ser.where([True, False, True], other)
 
+    def test_loc_new_category_raises(self):
+        ser = pd.Series(Categorical(["a", "b", "c"]))
+        msg = "Cannot setitem on a Categorical with a new category"
+        with pytest.raises(ValueError, match=msg):
+            ser.loc[3] = "d"
+
+    def test_unused_category_retention(self):
+        # Init case
+        exp_cats = Index(["a", "b", "c", "d"])
+        cat1 = Series(Categorical(["a", "b", "c"], categories=exp_cats))
+        tm.assert_index_equal(cat1.cat.categories, exp_cats)
+
+        # Modify case
+        cat1.loc[0] = "b"
+        tm.assert_index_equal(cat1.cat.categories, exp_cats)
+
 
 @pytest.mark.parametrize("index", [True, False])
 def test_mask_with_boolean(index):