ENH: Support mask in unique

asv_bench/benchmarks/hash_functions.py

@@ -39,6 +39,21 @@ def time_unique(self, exponent):
         pd.unique(self.a2)
 
 
+class Unique:
+    params = ["Int64", "Float64"]
+    param_names = ["dtype"]
+
+    def setup(self, dtype):
+        self.ser = pd.Series(([1, pd.NA, 2] + list(range(100_000))) * 3, dtype=dtype)
+        self.ser_unique = pd.Series(list(range(300_000)) + [pd.NA], dtype=dtype)
+
+    def time_unique_with_duplicates(self, exponent):
+        pd.unique(self.ser)
+
+    def time_unique(self, exponent):
+        pd.unique(self.ser_unique)
+
+
 class NumericSeriesIndexing:
 
     params = [

pandas/_libs/hashtable_class_helper.pxi.in

@@ -521,7 +521,7 @@ cdef class {{name}}HashTable(HashTable):
     def _unique(self, const {{dtype}}_t[:] values, {{name}}Vector uniques,
                 Py_ssize_t count_prior=0, Py_ssize_t na_sentinel=-1,
                 object na_value=None, bint ignore_na=False,
-                object mask=None, bint return_inverse=False):
+                object mask=None, bint return_inverse=False, bint use_result_mask=False):
         """
         Calculate unique values and labels (no sorting!)
 
@@ -551,6 +551,9 @@ cdef class {{name}}HashTable(HashTable):
         return_inverse : bool, default False
             Whether the mapping of the original array values to their location
             in the vector of uniques should be returned.
+        use_result_mask: bool, default False
+            Whether to create a result mask for the unique values. Not supported
+            with return_inverse=True.
 
         Returns
         -------
@@ -566,14 +569,24 @@ cdef class {{name}}HashTable(HashTable):
             {{c_type}} val, na_value2
             khiter_t k
             {{name}}VectorData *ud
-            bint use_na_value, use_mask
+            UInt8Vector result_mask
+            UInt8VectorData *rmd
+            bint use_na_value, use_mask, seen_na = False
             uint8_t[:] mask_values
 
         if return_inverse:
             labels = np.empty(n, dtype=np.intp)
         ud = uniques.data
         use_na_value = na_value is not None
         use_mask = mask is not None
+        if not use_mask and use_result_mask:
+            raise NotImplementedError  # pragma: no cover
+
+        if use_result_mask and return_inverse:
+            raise NotImplementedError  # pragma: no cover
+
+        result_mask = UInt8Vector()
+        rmd = result_mask.data
 
         if use_mask:
             mask_values = mask.view("uint8")
@@ -605,6 +618,27 @@ cdef class {{name}}HashTable(HashTable):
                     # and replace the corresponding label with na_sentinel
                     labels[i] = na_sentinel
                     continue
+                elif not ignore_na and use_result_mask:
+                    if mask_values[i]:
+                        if seen_na:
+                            continue
+
+                        seen_na = True
+                        if needs_resize(ud):
+                            with gil:
+                                if uniques.external_view_exists:
+                                    raise ValueError("external reference to "
+                                                     "uniques held, but "
+                                                     "Vector.resize() needed")
+                                uniques.resize()
+                                if result_mask.external_view_exists:
+                                    raise ValueError("external reference to "
+                                                     "result_mask held, but "
+                                                     "Vector.resize() needed")
+                                result_mask.resize()
+                        append_data_{{dtype}}(ud, val)
+                        append_data_uint8(rmd, 1)
+                        continue
 
                 k = kh_get_{{dtype}}(self.table, val)
 
@@ -619,7 +653,16 @@ cdef class {{name}}HashTable(HashTable):
                                                  "uniques held, but "
                                                  "Vector.resize() needed")
                             uniques.resize()
+                            if use_result_mask:
+                                if result_mask.external_view_exists:
+                                    raise ValueError("external reference to "
+                                                     "result_mask held, but "
+                                                     "Vector.resize() needed")
+                                result_mask.resize()
                     append_data_{{dtype}}(ud, val)
+                    if use_result_mask:
+                        append_data_uint8(rmd, 0)
+
                     if return_inverse:
                         self.table.vals[k] = count
                         labels[i] = count
@@ -632,9 +675,11 @@ cdef class {{name}}HashTable(HashTable):
 
         if return_inverse:
             return uniques.to_array(), labels.base  # .base -> underlying ndarray
+        if use_result_mask:
+            return uniques.to_array(), result_mask.to_array()
         return uniques.to_array()
 
-    def unique(self, const {{dtype}}_t[:] values, bint return_inverse=False):
+    def unique(self, const {{dtype}}_t[:] values, bint return_inverse=False, object mask=None):
         """
         Calculate unique values and labels (no sorting!)
 
@@ -652,10 +697,14 @@ cdef class {{name}}HashTable(HashTable):
             Unique values of input, not sorted
         labels : ndarray[intp_t] (if return_inverse)
             The labels from values to uniques
+        mask : ndarray[bool], optional
+            If not None, the mask is used as indicator for missing values
+            (True = missing, False = valid) instead of `na_value` or
         """
         uniques = {{name}}Vector()
+        use_result_mask = True if mask is not None else False
         return self._unique(values, uniques, ignore_na=False,
-                            return_inverse=return_inverse)
+                            return_inverse=return_inverse, mask=mask, use_result_mask=use_result_mask)
 
     def factorize(self, const {{dtype}}_t[:] values, Py_ssize_t na_sentinel=-1,
                   object na_value=None, object mask=None):
@@ -1013,7 +1062,7 @@ cdef class StringHashTable(HashTable):
             return uniques.to_array(), labels.base  # .base -> underlying ndarray
         return uniques.to_array()
 
-    def unique(self, ndarray[object] values, bint return_inverse=False):
+    def unique(self, ndarray[object] values, bint return_inverse=False, object mask=None):
         """
         Calculate unique values and labels (no sorting!)
 
@@ -1024,6 +1073,8 @@ cdef class StringHashTable(HashTable):
         return_inverse : bool, default False
             Whether the mapping of the original array values to their location
             in the vector of uniques should be returned.
+        mask : ndarray[bool], optional
+            Not yet implemented for StringHashTable
 
         Returns
         -------
@@ -1266,7 +1317,7 @@ cdef class PyObjectHashTable(HashTable):
             return uniques.to_array(), labels.base  # .base -> underlying ndarray
         return uniques.to_array()
 
-    def unique(self, ndarray[object] values, bint return_inverse=False):
+    def unique(self, ndarray[object] values, bint return_inverse=False, object mask=None):
         """
         Calculate unique values and labels (no sorting!)
 
@@ -1277,6 +1328,8 @@ cdef class PyObjectHashTable(HashTable):
         return_inverse : bool, default False
             Whether the mapping of the original array values to their location
             in the vector of uniques should be returned.
+        mask : ndarray[bool], optional
+            Not yet implemented for PyObjectHashTable
 
         Returns
         -------

pandas/core/algorithms.py

@@ -322,6 +322,8 @@ def unique(values):
     Parameters
     ----------
     values : 1d array-like
+    mask : 1d bool array
+        Mask of the values.
 
     Returns
     -------
@@ -404,6 +406,11 @@ def unique(values):
     >>> pd.unique([("a", "b"), ("b", "a"), ("a", "c"), ("b", "a")])
     array([('a', 'b'), ('b', 'a'), ('a', 'c')], dtype=object)
     """
+    return _unique(values)
+
+
+def _unique(values, mask: npt.NDArray[np.bool_] | None = None):
+    """See algorithms.unique for docs"""
     values = _ensure_arraylike(values)
 
     if is_extension_array_dtype(values.dtype):
@@ -414,9 +421,15 @@ def unique(values):
     htable, values = _get_hashtable_algo(values)
 
     table = htable(len(values))
-    uniques = table.unique(values)
-    uniques = _reconstruct_data(uniques, original.dtype, original)
-    return uniques
+    if mask is None:
+        uniques = table.unique(values)
+        uniques = _reconstruct_data(uniques, original.dtype, original)
+        return uniques
+
+    else:
+        uniques, mask = table.unique(values, mask=mask)
+        uniques = _reconstruct_data(uniques, original.dtype, original)
+        return uniques, mask.astype("bool")
 
 
 unique1d = unique

pandas/core/arrays/masked.py

@@ -851,6 +851,17 @@ def copy(self: BaseMaskedArrayT) -> BaseMaskedArrayT:
         mask = mask.copy()
         return type(self)(data, mask, copy=False)
 
+    def unique(self: BaseMaskedArrayT) -> BaseMaskedArrayT:
+        """
+        Compute the BaseMaskedArray of unique values.
+
+        Returns
+        -------
+        uniques : BaseMaskedArray
+        """
+        uniques, mask = algos._unique(self._data, self._mask)
+        return type(self)(uniques, mask, copy=False)
+
     @doc(ExtensionArray.searchsorted)
     def searchsorted(
         self,

pandas/tests/test_algos.py

@@ -834,6 +834,13 @@ def test_do_not_mangle_na_values(self, unique_nulls_fixture, unique_nulls_fixtur
         assert a[0] is unique_nulls_fixture
         assert a[1] is unique_nulls_fixture2
 
+    def test_unique_masked(self, any_numeric_ea_dtype):
+        # GH#48019
+        ser = Series([1, pd.NA, 2] * 3, dtype=any_numeric_ea_dtype)
+        result = pd.unique(ser)
+        expected = pd.array([1, pd.NA, 2], dtype=any_numeric_ea_dtype)
+        tm.assert_extension_array_equal(result, expected)
+
 
 class TestIsin:
     def test_invalid(self):