Adds explode API

python/cudf/cudf/_lib/cpp/lists/explode.pxd

@@ -0,0 +1,13 @@
+# Copyright (c) 2021, NVIDIA CORPORATION.
+
+from libcpp.memory cimport unique_ptr
+
+from cudf._lib.cpp.table.table cimport table
+from cudf._lib.cpp.table.table_view cimport table_view
+from cudf._lib.cpp.types cimport size_type
+
+cdef extern from "cudf/lists/explode.hpp" namespace "cudf" nogil:
+    cdef unique_ptr[table] explode_outer(
+        const table_view,
+        size_type explode_column_idx,
+    ) except +

python/cudf/cudf/_lib/lists.pyx

@@ -6,12 +6,19 @@ from libcpp.utility cimport move
 from cudf._lib.cpp.lists.count_elements cimport (
     count_elements as cpp_count_elements
 )
+from cudf._lib.cpp.lists.explode cimport (
+    explode_outer as cpp_explode_outer
+)
 from cudf._lib.cpp.lists.lists_column_view cimport lists_column_view
 from cudf._lib.cpp.column.column_view cimport column_view
 from cudf._lib.cpp.column.column cimport column
 
-from cudf._lib.column cimport Column
+from cudf._lib.cpp.table.table cimport table
+from cudf._lib.cpp.table.table_view cimport table_view
+from cudf._lib.cpp.types cimport size_type
 
+from cudf._lib.column cimport Column
+from cudf._lib.table cimport Table
 
 from cudf.core.dtypes import ListDtype
 
@@ -32,3 +39,19 @@ def count_elements(Column col):
 
     result = Column.from_unique_ptr(move(c_result))
     return result
+
+
+def explode_outer(Table tbl, int explode_column_idx):
+    cdef table_view c_table_view = tbl.view()
+    cdef size_type c_explode_column_idx = explode_column_idx
+
+    cdef unique_ptr[table] c_result
+
+    with nogil:
+        c_result = move(cpp_explode_outer(c_table_view, c_explode_column_idx))
+
+    return Table.from_unique_ptr(
+        move(c_result),
+        column_names=tbl._column_names,
+        index_names=tbl._index_names
+    )

python/cudf/cudf/core/dataframe.py

@@ -7709,6 +7709,50 @@ def equals(self, other):
                 return False
         return super().equals(other)
 
+    def explode(self, column, ignore_index=False):
+        """
+        Transform each element of a list-like to a row, replicating index
+        values.
+
+        Parameters
+        ----------
+        column : str or tuple
+            Column to explode.
+        ignore_index : bool, default False
+            If True, the resulting index will be labeled 0, 1, …, n - 1.
+
+        Returns
+        -------
+        DataFrame
+
+        Examples
+        -------
+        >>> import cudf
+        >>> cudf.DataFrame(
+                {"a": [[1, 2, 3], [], None, [4, 5]], "b": [11, 22, 33, 44]})
+                   a   b
+        0  [1, 2, 3]  11
+        1         []  22
+        2       None  33
+        3     [4, 5]  44
+        >>> df.explode('a')
+              a   b
+        0     1  11
+        0     2  11
+        0     3  11
+        1  <NA>  22
+        2  <NA>  33
+        3     4  44
+        3     5  44
+        """
+        if column not in self._column_names:
+            raise KeyError(column)
+
+        explode_num = self._column_names.index(column)
+        return super()._explode(
+            explode_num, None if ignore_index else self.index
+        )
+
     _accessors = set()  # type: Set[Any]
 
 

python/cudf/cudf/core/frame.py

@@ -6,7 +6,16 @@
 import functools
 import warnings
 from collections import OrderedDict, abc as abc
-from typing import TYPE_CHECKING, Any, Dict, Tuple, TypeVar, Union, overload
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    Dict,
+    Optional,
+    Tuple,
+    TypeVar,
+    Union,
+    overload,
+)
 
 import cupy
 import numpy as np
@@ -573,6 +582,18 @@ def equals(self, other, **kwargs):
         else:
             return self._index.equals(other._index)
 
+    def _explode(self, explode_column_num: int, index: Optional[cudf.Index]):
+        if index is not None:
+            explode_column_num += index.nlevels
+        res_tbl = libcudf.lists.explode_outer(
+            cudf._lib.table.Table(self._data, index=index), explode_column_num
+        )
+
+        res = self.__class__._from_table(res_tbl)
+        if index is not None:
+            res.index.names = index.names
+        return res
+
     def _get_columns_by_label(self, labels, downcast):
         """
         Returns columns of the Frame specified by `labels`

python/cudf/cudf/core/series.py

@@ -6364,6 +6364,43 @@ def keys(self):
         """
         return self.index
 
+    def explode(self, ignore_index=False):
+        """
+        Transform each element of a list-like to a row, replicating index
+        values.
+
+        Parameters
+        ----------
+        ignore_index : bool, default False
+            If True, the resulting index will be labeled 0, 1, …, n - 1.
+
+        Returns
+        -------
+        DataFrame
+
+        Examples
+        -------
+        >>> import cudf
+        >>> s = cudf.Series([[1, 2, 3], [], None, [4, 5]])
+        >>> s
+        0    [1, 2, 3]
+        1           []
+        2         None
+        3       [4, 5]
+        dtype: list
+        >>> s.explode()
+        0       1
+        0       2
+        0       3
+        1    <NA>
+        2    <NA>
+        3       4
+        3       5
+        dtype: int64
+        """
+
+        return super()._explode(0, None if ignore_index else self.index)
+
     _accessors = set()  # type: Set[Any]
 
 

python/cudf/cudf/tests/test_dataframe.py

@@ -8442,3 +8442,31 @@ def test_rename_for_level_is_None_MC():
     got = gdf.rename(columns={"a": "f"}, level=None)
 
     assert_eq(expect, got)
+
+
+@pytest.mark.parametrize("ignore_index", [True, False])
+@pytest.mark.parametrize(
+    "p_index",
+    [
+        None,
+        ["ia", "ib", "ic", "id", "ie"],
+        pd.MultiIndex.from_tuples(
+            [(0, "a"), (0, "b"), (0, "c"), (1, "a"), (1, "b")]
+        ),
+    ],
+)
+def test_explode(ignore_index, p_index):
+    gdf = cudf.DataFrame(
+        {
+            "a": [[1, 2, 3], None, [4], [], [5, 6]],
+            "b": [11, 22, 33, 44, 55],
+            "c": ["a", "e", "i", "o", "u"],
+        },
+        index=p_index,
+    )
+    pdf = gdf.to_pandas(nullable=True)
+
+    expect = pdf.explode("a", ignore_index)
+    got = gdf.explode("a", ignore_index)
+
+    assert_eq(got, expect, check_dtype=False)

python/cudf/cudf/tests/test_series.py

@@ -1118,3 +1118,24 @@ def test_series_drop_raises():
     actual = gs.drop("p", errors="ignore")
 
     assert_eq(actual, expect)
+
+
+@pytest.mark.parametrize("ignore_index", [True, False])
+@pytest.mark.parametrize(
+    "p_index",
+    [
+        None,
+        ["ia", "ib", "ic", "id", "ie"],
+        pd.MultiIndex.from_tuples(
+            [(0, "a"), (0, "b"), (0, "c"), (1, "a"), (1, "b")]
+        ),
+    ],
+)
+def test_explode(ignore_index, p_index):
+    gdf = cudf.Series([[1, 2, 3], None, [4], [], [5, 6]], index=p_index)
+    pdf = gdf.to_pandas(nullable=True)
+
+    expect = pdf.explode(ignore_index)
+    got = gdf.explode(ignore_index)
+
+    assert_eq(expect, got, check_dtype=False)