Expose seed argument to hash_values

python/cudf/cudf/core/indexed_frame.py

@@ -1629,7 +1629,7 @@ def memory_usage(self, index=True, deep=False):
         """
         raise NotImplementedError
 
-    def hash_values(self, method="murmur3"):
+    def hash_values(self, method="murmur3", seed=None):
         """Compute the hash of values in this column.
 
         Parameters
@@ -1639,6 +1639,12 @@ def hash_values(self, method="murmur3"):
             * murmur3: MurmurHash3 hash function.
             * md5: MD5 hash function.
 
+        seed : int, optional
+            Seed value to use for the hash function.
+            Note - This only has effect for the following supported
+            hash functions:
+            * murmur3: MurmurHash3 hash function.
+
         Returns
         -------
         Series
@@ -1665,6 +1671,11 @@ def hash_values(self, method="murmur3"):
         1    947ca8d2c5f0f27437f156cfbfab0969
         2    d0580ef52d27c043c8e341fd5039b166
         dtype: object
+        >>> series.hash_values(method="murmur3", seed=42)
+        0    2364453205
+        1     422621911
+        2    3353449140
+        dtype: uint32
 
         **DataFrame**
 
@@ -1686,11 +1697,20 @@ def hash_values(self, method="murmur3"):
         2    fe061786ea286a515b772d91b0dfcd70
         dtype: object
         """
+        seed_hash_methods = {"murmur3"}
+        if seed is None:
+            seed = 0
+        elif method not in seed_hash_methods:
+            warnings.warn(
+                "Provided seed value has no effect for hash method"
+                f" `{method}`. Refer to the docstring for information"
+                " on hash methods that support the `seed` param"
+            )
         # Note that both Series and DataFrame return Series objects from this
         # calculation, necessitating the unfortunate circular reference to the
         # child class here.
         return cudf.Series._from_data(
-            {None: libcudf.hash.hash([*self._columns], method)},
+            {None: libcudf.hash.hash([*self._columns], method, seed)},
             index=self.index,
         )
 

python/cudf/cudf/tests/test_dataframe.py

@@ -38,6 +38,7 @@
     NUMERIC_TYPES,
     assert_eq,
     assert_exceptions_equal,
+    assert_neq,
     does_not_raise,
     expect_warning_if,
     gen_rand,
@@ -1323,9 +1324,10 @@ def test_assign():
 
 @pytest.mark.parametrize("nrows", [1, 8, 100, 1000])
 @pytest.mark.parametrize("method", ["murmur3", "md5"])
-def test_dataframe_hash_values(nrows, method):
+@pytest.mark.parametrize("seed", [None, 42])
+def test_dataframe_hash_values(nrows, method, seed):
     gdf = cudf.DataFrame()
-    data = np.asarray(range(nrows))
+    data = np.arange(nrows)
     data[0] = data[-1]  # make first and last the same
     gdf["a"] = data
     gdf["b"] = gdf.a + 100
@@ -1334,12 +1336,41 @@ def test_dataframe_hash_values(nrows, method):
     assert len(out) == nrows
     assert out.dtype == np.uint32
 
+    warning_expected = (
+        True if seed is not None and method not in {"murmur3"} else False
+    )
     # Check single column
-    out_one = gdf[["a"]].hash_values(method=method)
+    if warning_expected:
+        with pytest.warns(
+            UserWarning, match="Provided seed value has no effect*"
+        ):
+            out_one = gdf[["a"]].hash_values(method=method, seed=seed)
+    else:
+        out_one = gdf[["a"]].hash_values(method=method, seed=seed)
     # First matches last
     assert out_one.iloc[0] == out_one.iloc[-1]
     # Equivalent to the cudf.Series.hash_values()
-    assert_eq(gdf["a"].hash_values(method=method), out_one)
+    if warning_expected:
+        with pytest.warns(
+            UserWarning, match="Provided seed value has no effect*"
+        ):
+            assert_eq(gdf["a"].hash_values(method=method, seed=seed), out_one)
+    else:
+        assert_eq(gdf["a"].hash_values(method=method, seed=seed), out_one)
+
+
+@pytest.mark.parametrize("method", ["murmur3"])
+def test_dataframe_hash_values_seed(method):
+    gdf = cudf.DataFrame()
+    data = np.arange(10)
+    data[0] = data[-1]  # make first and last the same
+    gdf["a"] = data
+    gdf["b"] = gdf.a + 100
+    out_one = gdf.hash_values(method=method, seed=0)
+    out_two = gdf.hash_values(method=method, seed=1)
+    assert out_one.iloc[0] == out_one.iloc[-1]
+    assert out_two.iloc[0] == out_two.iloc[-1]
+    assert_neq(out_one, out_two)
 
 
 @pytest.mark.parametrize("nrows", [3, 10, 100, 1000])