Implement DataFrame diff() (#9817)

Fixes: #9604 and resolves #1271

Authors:
  - Sheilah Kirui (https://github.com/skirui-source)

Approvers:
  - Karthikeyan (https://github.com/karthikeyann)
  - Bradley Dice (https://github.com/bdice)
  - Michael Wang (https://github.com/isVoid)

URL: #9817

python/cudf/cudf/core/dataframe.py

@@ -19,6 +19,7 @@
 import pyarrow as pa
 from nvtx import annotate
 from pandas._config import get_option
+from pandas.core.dtypes.common import is_float, is_integer
 from pandas.io.formats import console
 from pandas.io.formats.printing import pprint_thing
 
@@ -2542,6 +2543,80 @@ def insert(self, loc, name, value, nan_as_null=None):
 
         self._data.insert(name, value, loc=loc)
 
+    def diff(self, periods=1, axis=0):
+        """
+        First discrete difference of element.
+
+        Calculates the difference of a DataFrame element compared with another
+        element in the DataFrame (default is element in previous row).
+
+        Parameters
+        ----------
+        periods : int, default 1
+            Periods to shift for calculating difference,
+            accepts negative values.
+        axis : {0 or 'index', 1 or 'columns'}, default 0
+            Take difference over rows (0) or columns (1).
+            Only row-wise (0) shift is supported.
+
+        Returns
+        -------
+        DataFrame
+            First differences of the DataFrame.
+
+        Notes
+        -----
+        Diff currently only supports numeric dtype columns.
+
+        Examples
+        --------
+        >>> import cudf
+        >>> gdf = cudf.DataFrame({'a': [1, 2, 3, 4, 5, 6],
+        ...                       'b': [1, 1, 2, 3, 5, 8],
+        ...                       'c': [1, 4, 9, 16, 25, 36]})
+        >>> gdf
+           a  b   c
+        0  1  1   1
+        1  2  1   4
+        2  3  2   9
+        3  4  3  16
+        4  5  5  25
+        5  6  8  36
+        >>> gdf.diff(periods=2)
+              a     b     c
+        0  <NA>  <NA>  <NA>
+        1  <NA>  <NA>  <NA>
+        2     2     1     8
+        3     2     2    12
+        4     2     3    16
+        5     2     5    20
+
+        """
+        if not is_integer(periods):
+            if not (is_float(periods) and periods.is_integer()):
+                raise ValueError("periods must be an integer")
+            periods = int(periods)
+
+        axis = self._get_axis_from_axis_arg(axis)
+        if axis != 0:
+            raise NotImplementedError("Only axis=0 is supported.")
+
+        if not all(is_numeric_dtype(i) for i in self.dtypes):
+            raise NotImplementedError(
+                "DataFrame.diff only supports numeric dtypes"
+            )
+
+        if abs(periods) > len(self):
+            df = cudf.DataFrame._from_data(
+                {
+                    name: column_empty(len(self), dtype=dtype, masked=True)
+                    for name, dtype in zip(self.columns, self.dtypes)
+                }
+            )
+            return df
+
+        return self - self.shift(periods=periods)
+
     def drop(
         self,
         labels=None,

python/cudf/cudf/tests/test_dataframe.py

@@ -9065,6 +9065,77 @@ def test_dataframe_add_suffix():
     assert_eq(got, expected)
 
 
+@pytest.mark.parametrize(
+    "data",
+    [
+        np.random.RandomState(seed=10).randint(-50, 50, (25, 30)),
+        np.random.RandomState(seed=10).random_sample((4, 4)),
+        np.array([1.123, 2.343, 5.890, 0.0]),
+        [True, False, True, False, False],
+        {"a": [1.123, 2.343, np.nan, np.nan], "b": [None, 3, 9.08, None]},
+    ],
+)
+@pytest.mark.parametrize("periods", (-5, -1, 0, 1, 5))
+def test_diff_dataframe_numeric_dtypes(data, periods):
+    gdf = cudf.DataFrame(data)
+    pdf = gdf.to_pandas()
+
+    actual = gdf.diff(periods=periods, axis=0)
+    expected = pdf.diff(periods=periods, axis=0)
+
+    assert_eq(
+        expected, actual, check_dtype=False,
+    )
+
+
+@pytest.mark.parametrize(
+    ("precision", "scale"), [(5, 2), (8, 5)],
+)
+@pytest.mark.parametrize(
+    "dtype", [cudf.Decimal32Dtype, cudf.Decimal64Dtype],
+)
+def test_diff_decimal_dtypes(precision, scale, dtype):
+    gdf = cudf.DataFrame(
+        np.random.default_rng(seed=42).uniform(10.5, 75.5, (10, 6)),
+        dtype=dtype(precision=precision, scale=scale),
+    )
+    pdf = gdf.to_pandas()
+
+    actual = gdf.diff()
+    expected = pdf.diff()
+
+    assert_eq(
+        expected, actual, check_dtype=False,
+    )
+
+
+def test_diff_dataframe_invalid_axis():
+    gdf = cudf.DataFrame(np.array([1.123, 2.343, 5.890, 0.0]))
+    with pytest.raises(NotImplementedError, match="Only axis=0 is supported."):
+        gdf.diff(periods=1, axis=1)
+
+
+@pytest.mark.parametrize(
+    "data",
+    [
+        {
+            "int_col": [1, 2, 3, 4, 5],
+            "float_col": [1.0, 2.0, 3.0, 4.0, 5.0],
+            "string_col": ["a", "b", "c", "d", "e"],
+        },
+        ["a", "b", "c", "d", "e"],
+        [np.nan, None, np.nan, None],
+    ],
+)
+def test_diff_dataframe_non_numeric_dypes(data):
+    gdf = cudf.DataFrame(data)
+    with pytest.raises(
+        NotImplementedError,
+        match="DataFrame.diff only supports numeric dtypes",
+    ):
+        gdf.diff(periods=2, axis=0)
+
+
 def test_dataframe_assign_cp_np_array():
     m, n = 5, 3
     cp_ndarray = cupy.random.randn(m, n)