Merge pull request #2595 from IntelPython/reuse-tensor-isin

Implement `dpnp.isin`

Author: ndgrigorian

CHANGELOG.md

@@ -25,6 +25,7 @@ Also, that release drops support for Python 3.9, making Python 3.10 the minimum
 * Added implementation of `dpnp.ndarray.__format__` method [#2662](https://github.com/IntelPython/dpnp/pull/2662)
 * Added implementation of `dpnp.ndarray.__bytes__` method [#2671](https://github.com/IntelPython/dpnp/pull/2671)
 * Added implementation of `dpnp.divmod` [#2674](https://github.com/IntelPython/dpnp/pull/2674)
+* Added implementation of `dpnp.isin` function [#2595](https://github.com/IntelPython/dpnp/pull/2595)
 
 ### Changed
 

dpnp/__init__.py

@@ -498,6 +498,7 @@
     unique_inverse,
     unique_values,
 )
+from .dpnp_iface_logic import isin
 
 # -----------------------------------------------------------------------------
 # Sorting, searching, and counting
@@ -981,6 +982,7 @@
 
 # Set routines
 __all__ += [
+    "isin",
     "unique",
     "unique_all",
     "unique_counts",

dpnp/dpnp_iface_logic.py

@@ -53,6 +53,7 @@
 import dpnp.backend.extensions.ufunc._ufunc_impl as ufi
 from dpnp.dpnp_algo.dpnp_elementwise_common import DPNPBinaryFunc, DPNPUnaryFunc
 
+from .dpnp_array import dpnp_array
 from .dpnp_utils import get_usm_allocations
 
 
@@ -1166,6 +1167,120 @@ def isfortran(a):
     return a.flags.fnc
 
 
+def isin(
+    element,
+    test_elements,
+    assume_unique=False,  # pylint: disable=unused-argument
+    invert=False,
+    *,
+    kind=None,  # pylint: disable=unused-argument
+):
+    """
+    Calculates ``element in test_elements``, broadcasting over `element` only.
+    Returns a boolean array of the same shape as `element` that is ``True``
+    where an element of `element` is in `test_elements` and ``False``
+    otherwise.
+
+    For full documentation refer to :obj:`numpy.isin`.
+
+    Parameters
+    ----------
+    element : {dpnp.ndarray, usm_ndarray, scalar}
+        Input array.
+    test_elements : {dpnp.ndarray, usm_ndarray, scalar}
+        The values against which to test each value of `element`.
+        This argument is flattened if it is an array.
+    assume_unique : bool, optional
+        Ignored, as no performance benefit is gained by assuming the
+        input arrays are unique. Included for compatibility with NumPy.
+
+        Default: ``False``.
+    invert : bool, optional
+        If ``True``, the values in the returned array are inverted, as if
+        calculating ``element not in test_elements``.
+        ``dpnp.isin(a, b, invert=True)`` is equivalent to (but faster
+        than) ``dpnp.invert(dpnp.isin(a, b))``.
+
+        Default: ``False``.
+    kind : {None, "sort"}, optional
+        Ignored, as the only algorithm implemented is ``"sort"``. Included for
+        compatibility with NumPy.
+
+        Default: ``None``.
+
+    Returns
+    -------
+    isin : dpnp.ndarray of bool dtype
+        Has the same shape as `element`. The values `element[isin]`
+        are in `test_elements`.
+
+    Examples
+    --------
+    >>> import dpnp as np
+    >>> element = 2*np.arange(4).reshape((2, 2))
+    >>> element
+    array([[0, 2],
+           [4, 6]])
+    >>> test_elements = np.array([1, 2, 4, 8])
+    >>> mask = np.isin(element, test_elements)
+    >>> mask
+    array([[False,  True],
+           [ True, False]])
+    >>> element[mask]
+    array([2, 4])
+
+    The indices of the matched values can be obtained with `nonzero`:
+
+    >>> np.nonzero(mask)
+    (array([0, 1]), array([1, 0]))
+
+    The test can also be inverted:
+
+    >>> mask = np.isin(element, test_elements, invert=True)
+    >>> mask
+    array([[ True, False],
+           [False,  True]])
+    >>> element[mask]
+    array([0, 6])
+
+    """
+
+    dpnp.check_supported_arrays_type(element, test_elements, scalar_type=True)
+    if dpnp.isscalar(element):
+        usm_element = dpnp.as_usm_ndarray(
+            element,
+            usm_type=test_elements.usm_type,
+            sycl_queue=test_elements.sycl_queue,
+        )
+        usm_test = dpnp.get_usm_ndarray(test_elements)
+    elif dpnp.isscalar(test_elements):
+        usm_test = dpnp.as_usm_ndarray(
+            test_elements,
+            usm_type=element.usm_type,
+            sycl_queue=element.sycl_queue,
+        )
+        usm_element = dpnp.get_usm_ndarray(element)
+    else:
+        if (
+            dpu.get_execution_queue(
+                (element.sycl_queue, test_elements.sycl_queue)
+            )
+            is None
+        ):
+            raise dpu.ExecutionPlacementError(
+                "Input arrays have incompatible allocation queues"
+            )
+        usm_element = dpnp.get_usm_ndarray(element)
+        usm_test = dpnp.get_usm_ndarray(test_elements)
+    return dpnp_array._create_from_usm_ndarray(
+        dpt.isin(
+            usm_element,
+            usm_test,
+            invert=invert,
+        )
+    )
+
+
 _ISINF_DOCSTRING = """
 Tests each element :math:`x_i` of the input array `x` to determine if equal to
 positive or negative infinity.

dpnp/tests/test_logic.py

@@ -1,5 +1,7 @@
+import dpctl
 import numpy
 import pytest
+from dpctl.utils import ExecutionPlacementError
 from numpy.testing import (
     assert_allclose,
     assert_array_equal,
@@ -795,3 +797,103 @@ def test_array_equal_nan(a):
     result = dpnp.array_equal(dpnp.array(a), dpnp.array(b), equal_nan=True)
     expected = numpy.array_equal(a, b, equal_nan=True)
     assert_equal(result, expected)
+
+
+class TestIsin:
+    @pytest.mark.parametrize(
+        "a",
+        [
+            numpy.array([1, 2, 3, 4]),
+            numpy.array([[1, 2], [3, 4]]),
+        ],
+    )
+    @pytest.mark.parametrize(
+        "b",
+        [
+            numpy.array([2, 4, 6]),
+            numpy.array([[1, 3], [5, 7]]),
+        ],
+    )
+    def test_isin_basic(self, a, b):
+        dp_a = dpnp.array(a)
+        dp_b = dpnp.array(b)
+
+        expected = numpy.isin(a, b)
+        result = dpnp.isin(dp_a, dp_b)
+        assert_equal(result, expected)
+
+    @pytest.mark.parametrize("dtype", get_all_dtypes(no_none=True))
+    def test_isin_dtype(self, dtype):
+        a = numpy.array([1, 2, 3, 4], dtype=dtype)
+        b = numpy.array([2, 4], dtype=dtype)
+
+        dp_a = dpnp.array(a, dtype=dtype)
+        dp_b = dpnp.array(b, dtype=dtype)
+
+        expected = numpy.isin(a, b)
+        result = dpnp.isin(dp_a, dp_b)
+        assert_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "sh_a, sh_b", [((3, 1), (1, 4)), ((2, 3, 1), (1, 1))]
+    )
+    def test_isin_broadcast(self, sh_a, sh_b):
+        a = numpy.arange(numpy.prod(sh_a)).reshape(sh_a)
+        b = numpy.arange(numpy.prod(sh_b)).reshape(sh_b)
+
+        dp_a = dpnp.array(a)
+        dp_b = dpnp.array(b)
+
+        expected = numpy.isin(a, b)
+        result = dpnp.isin(dp_a, dp_b)
+        assert_equal(result, expected)
+
+    def test_isin_scalar_elements(self):
+        a = numpy.array([1, 2, 3])
+        b = 2
+
+        dp_a = dpnp.array(a)
+        dp_b = dpnp.array(b)
+
+        expected = numpy.isin(a, b)
+        result = dpnp.isin(dp_a, dp_b)
+        assert_equal(result, expected)
+
+    def test_isin_scalar_test_elements(self):
+        a = 2
+        b = numpy.array([1, 2, 3])
+
+        dp_a = dpnp.array(a)
+        dp_b = dpnp.array(b)
+
+        expected = numpy.isin(a, b)
+        result = dpnp.isin(dp_a, dp_b)
+        assert_equal(result, expected)
+
+    def test_isin_empty(self):
+        a = numpy.array([], dtype=int)
+        b = numpy.array([1, 2, 3])
+
+        dp_a = dpnp.array(a)
+        dp_b = dpnp.array(b)
+
+        expected = numpy.isin(a, b)
+        result = dpnp.isin(dp_a, dp_b)
+        assert_equal(result, expected)
+
+    def test_isin_errors(self):
+        q1 = dpctl.SyclQueue()
+        q2 = dpctl.SyclQueue()
+
+        a = dpnp.arange(5, sycl_queue=q1)
+        b = dpnp.arange(3, sycl_queue=q2)
+
+        # unsupported type for elements or test_elements
+        with pytest.raises(TypeError):
+            dpnp.isin(dict(), a)
+
+        with pytest.raises(TypeError):
+            dpnp.isin(a, dict())
+
+        with pytest.raises(ExecutionPlacementError):
+            dpnp.isin(a, b)

dpnp/tests/test_sycl_queue.py

@@ -564,6 +564,7 @@ def test_logic_op_1in(op, device):
         "greater",
         "greater_equal",
         "isclose",
+        "isin",
         "less",
         "less_equal",
         "logical_and",

dpnp/tests/test_usm_type.py

@@ -355,6 +355,7 @@ def test_logic_op_1in(op, usm_type_x):
         "greater",
         "greater_equal",
         "isclose",
+        "isin",
         "less",
         "less_equal",
         "logical_and",

dpnp/tests/third_party/cupy/logic_tests/test_truth.py

@@ -89,7 +89,6 @@ def test_with_out(self, xp, dtype):
         return out
 
 
-@pytest.mark.skip("isin() is not supported yet")
 @testing.parameterize(
     *testing.product(
         {