Migrate remaining nvtext NGrams APIs to pylibcudf (#17070)

Apart of #15162

Authors:
  - Matthew Murray (https://github.com/Matt711)

Approvers:
  - https://github.com/brandon-b-miller

URL: #17070

docs/cudf/source/user_guide/api_docs/pylibcudf/nvtext/index.rst

@@ -8,3 +8,4 @@ nvtext
     generate_ngrams
     jaccard
     minhash
+    ngrams_tokenize

docs/cudf/source/user_guide/api_docs/pylibcudf/nvtext/ngrams_tokenize.rst

@@ -0,0 +1,6 @@
+===============
+ngrams_tokenize
+===============
+
+.. automodule:: pylibcudf.nvtext.ngrams_tokenize
+   :members:

python/cudf/cudf/_lib/nvtext/ngrams_tokenize.pyx

@@ -2,48 +2,22 @@
 
 from cudf.core.buffer import acquire_spill_lock
 
-from libcpp.memory cimport unique_ptr
-from libcpp.utility cimport move
-
-from pylibcudf.libcudf.column.column cimport column
-from pylibcudf.libcudf.column.column_view cimport column_view
-from pylibcudf.libcudf.nvtext.ngrams_tokenize cimport (
-    ngrams_tokenize as cpp_ngrams_tokenize,
-)
-from pylibcudf.libcudf.scalar.scalar cimport string_scalar
-from pylibcudf.libcudf.types cimport size_type
-
 from cudf._lib.column cimport Column
-from cudf._lib.scalar cimport DeviceScalar
+
+from pylibcudf import nvtext
 
 
 @acquire_spill_lock()
 def ngrams_tokenize(
-    Column strings,
+    Column input,
     int ngrams,
     object py_delimiter,
     object py_separator
 ):
-
-    cdef DeviceScalar delimiter = py_delimiter.device_value
-    cdef DeviceScalar separator = py_separator.device_value
-
-    cdef column_view c_strings = strings.view()
-    cdef size_type c_ngrams = ngrams
-    cdef const string_scalar* c_separator = <const string_scalar*>separator\
-        .get_raw_ptr()
-    cdef const string_scalar* c_delimiter = <const string_scalar*>delimiter\
-        .get_raw_ptr()
-    cdef unique_ptr[column] c_result
-
-    with nogil:
-        c_result = move(
-            cpp_ngrams_tokenize(
-                c_strings,
-                c_ngrams,
-                c_delimiter[0],
-                c_separator[0]
-            )
-        )
-
-    return Column.from_unique_ptr(move(c_result))
+    result = nvtext.ngrams_tokenize.ngrams_tokenize(
+        input.to_pylibcudf(mode="read"),
+        ngrams,
+        py_delimiter.device_value.c_value,
+        py_separator.device_value.c_value
+    )
+    return Column.from_pylibcudf(result)

python/pylibcudf/pylibcudf/nvtext/CMakeLists.txt

@@ -12,7 +12,9 @@
 # the License.
 # =============================================================================
 
-set(cython_sources edit_distance.pyx generate_ngrams.pyx jaccard.pyx minhash.pyx)
+set(cython_sources edit_distance.pyx generate_ngrams.pyx jaccard.pyx minhash.pyx
+                   ngrams_tokenize.pyx
+)
 
 set(linked_libraries cudf::cudf)
 rapids_cython_create_modules(

python/pylibcudf/pylibcudf/nvtext/__init__.pxd

@@ -1,10 +1,17 @@
 # Copyright (c) 2024, NVIDIA CORPORATION.
 
-from . cimport edit_distance, generate_ngrams, jaccard, minhash
+from . cimport (
+    edit_distance,
+    generate_ngrams,
+    jaccard,
+    minhash,
+    ngrams_tokenize,
+)
 
 __all__ = [
     "edit_distance",
     "generate_ngrams",
     "jaccard",
-    "minhash"
+    "minhash",
+    "ngrams_tokenize"
 ]

python/pylibcudf/pylibcudf/nvtext/__init__.py

@@ -1,10 +1,11 @@
 # Copyright (c) 2024, NVIDIA CORPORATION.
 
-from . import edit_distance, generate_ngrams, jaccard, minhash
+from . import edit_distance, generate_ngrams, jaccard, minhash, ngrams_tokenize
 
 __all__ = [
     "edit_distance",
     "generate_ngrams",
     "jaccard",
     "minhash",
+    "ngrams_tokenize",
 ]

python/pylibcudf/pylibcudf/nvtext/ngrams_tokenize.pxd

@@ -0,0 +1,13 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from pylibcudf.column cimport Column
+from pylibcudf.libcudf.types cimport size_type
+from pylibcudf.scalar cimport Scalar
+
+
+cpdef Column ngrams_tokenize(
+    Column input,
+    size_type ngrams,
+    Scalar delimiter,
+    Scalar separator
+)

python/pylibcudf/pylibcudf/nvtext/ngrams_tokenize.pyx

@@ -0,0 +1,54 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+from cython.operator cimport dereference
+from libcpp.memory cimport unique_ptr
+from libcpp.utility cimport move
+from pylibcudf.column cimport Column
+from pylibcudf.libcudf.column.column cimport column
+from pylibcudf.libcudf.nvtext.ngrams_tokenize cimport (
+    ngrams_tokenize as cpp_ngrams_tokenize,
+)
+from pylibcudf.libcudf.scalar.scalar cimport string_scalar
+from pylibcudf.libcudf.types cimport size_type
+from pylibcudf.scalar cimport Scalar
+
+
+cpdef Column ngrams_tokenize(
+    Column input,
+    size_type ngrams,
+    Scalar delimiter,
+    Scalar separator
+):
+    """
+    Returns a single column of strings by tokenizing the input strings column
+    and then producing ngrams of each string.
+
+    For details, see :cpp:func:`ngrams_tokenize`
+
+    Parameters
+    ----------
+    input : Column
+        Input strings
+    ngrams : size_type
+        The ngram number to generate
+    delimiter : Scalar
+        UTF-8 characters used to separate each string into tokens.
+        An empty string will separate tokens using whitespace.
+    separator : Scalar
+        The string to use for separating ngram tokens
+
+    Returns
+    -------
+    Column
+        New strings columns of tokens
+    """
+    cdef unique_ptr[column] c_result
+
+    with nogil:
+        c_result = cpp_ngrams_tokenize(
+            input.view(),
+            ngrams,
+            dereference(<const string_scalar*>delimiter.get()),
+            dereference(<const string_scalar*>separator.get()),
+        )
+    return Column.from_libcudf(move(c_result))

python/pylibcudf/pylibcudf/tests/test_nvtext_ngrams_tokenize.py

@@ -0,0 +1,37 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+import pyarrow as pa
+import pylibcudf as plc
+import pytest
+from utils import assert_column_eq
+
+
+@pytest.fixture(scope="module")
+def input_col():
+    arr = ["a*b*c*d", "a b c d", "a-b-c-d", "a*b c-d"]
+    return pa.array(arr)
+
+
+@pytest.mark.parametrize("ngrams", [2, 3])
+@pytest.mark.parametrize("delim", ["*", " ", "-"])
+@pytest.mark.parametrize("sep", ["_", "&", ","])
+def test_ngrams_tokenize(input_col, ngrams, delim, sep):
+    def ngrams_tokenize(strings, ngrams, delim, sep):
+        tokens = []
+        for s in strings:
+            ss = s.split(delim)
+            for i in range(len(ss) - ngrams + 1):
+                token = sep.join(ss[i : i + ngrams])
+                tokens.append(token)
+        return tokens
+
+    result = plc.nvtext.ngrams_tokenize.ngrams_tokenize(
+        plc.interop.from_arrow(input_col),
+        ngrams,
+        plc.interop.from_arrow(pa.scalar(delim)),
+        plc.interop.from_arrow(pa.scalar(sep)),
+    )
+    expected = pa.array(
+        ngrams_tokenize(input_col.to_pylist(), ngrams, delim, sep)
+    )
+    assert_column_eq(result, expected)