Implement dpnp.nan_to_num() (#1966)

* Implement dpnp.nan_to_num()

* Update cupy tests for nan_to_num()

* Add dpnp tests

* Skip test_nan_to_num_scalar_nan

* Applied review comments

* Add more tests for nan_to_num()

* Improve perfomance using out empty_like array

* Add checks for nan, posinf, neginf args

* Add type check for nan, posinf and neginf

* Update tests

* Add support boolean type

Author: vlad-perevezentsev

dpnp/dpnp_iface_mathematical.py

@@ -110,6 +110,7 @@
     "mod",
     "modf",
     "multiply",
+    "nan_to_num",
     "negative",
     "nextafter",
     "positive",
@@ -130,6 +131,13 @@
 ]
 
 
+def _get_max_min(dtype):
+    """Get the maximum and minimum representable values for an inexact dtype."""
+
+    f = dpnp.finfo(dtype)
+    return f.max, f.min
+
+
 def _get_reduction_res_dt(a, dtype, _out):
     """Get a data type used by dpctl for result array in reduction function."""
 
@@ -2353,6 +2361,141 @@ def modf(x1, **kwargs):
 )
 
 
+def nan_to_num(x, copy=True, nan=0.0, posinf=None, neginf=None):
+    """
+    Replace ``NaN`` with zero and infinity with large finite numbers (default
+    behaviour) or with the numbers defined by the user using the `nan`,
+    `posinf` and/or `neginf` keywords.
+
+    If `x` is inexact, ``NaN`` is replaced by zero or by the user defined value
+    in `nan` keyword, infinity is replaced by the largest finite floating point
+    values representable by ``x.dtype`` or by the user defined value in
+    `posinf` keyword and -infinity is replaced by the most negative finite
+    floating point values representable by ``x.dtype`` or by the user defined
+    value in `neginf` keyword.
+
+    For complex dtypes, the above is applied to each of the real and
+    imaginary components of `x` separately.
+
+    If `x` is not inexact, then no replacements are made.
+
+    For full documentation refer to :obj:`numpy.nan_to_num`.
+
+    Parameters
+    ----------
+    x : {dpnp.ndarray, usm_ndarray}
+        Input data.
+    copy : bool, optional
+        Whether to create a copy of `x` (``True``) or to replace values
+        in-place (``False``). The in-place operation only occurs if casting to
+        an array does not require a copy.
+    nan : {int, float, bool}, optional
+        Value to be used to fill ``NaN`` values.
+        Default: ``0.0``.
+    posinf : {int, float, bool, None}, optional
+        Value to be used to fill positive infinity values. If no value is
+        passed then positive infinity values will be replaced with a very
+        large number.
+        Default: ``None``.
+    neginf : {int, float, bool, None} optional
+        Value to be used to fill negative infinity values. If no value is
+        passed then negative infinity values will be replaced with a very
+        small (or negative) number.
+        Default: ``None``.
+
+    Returns
+    -------
+    out : dpnp.ndarray
+        `x`, with the non-finite values replaced. If `copy` is ``False``, this
+        may be `x` itself.
+
+    See Also
+    --------
+    :obj:`dpnp.isinf` : Shows which elements are positive or negative infinity.
+    :obj:`dpnp.isneginf` : Shows which elements are negative infinity.
+    :obj:`dpnp.isposinf` : Shows which elements are positive infinity.
+    :obj:`dpnp.isnan` : Shows which elements are Not a Number (NaN).
+    :obj:`dpnp.isfinite` : Shows which elements are finite
+                           (not NaN, not infinity)
+
+    Examples
+    --------
+    >>> import dpnp as np
+    >>> np.nan_to_num(np.array(np.inf))
+    array(1.79769313e+308)
+    >>> np.nan_to_num(np.array(-np.inf))
+    array(-1.79769313e+308)
+    >>> np.nan_to_num(np.array(np.nan))
+    array(0.)
+    >>> x = np.array([np.inf, -np.inf, np.nan, -128, 128])
+    >>> np.nan_to_num(x)
+    array([ 1.79769313e+308, -1.79769313e+308,  0.00000000e+000,
+           -1.28000000e+002,  1.28000000e+002])
+    >>> np.nan_to_num(x, nan=-9999, posinf=33333333, neginf=33333333)
+    array([ 3.3333333e+07,  3.3333333e+07, -9.9990000e+03, -1.2800000e+02,
+            1.2800000e+02])
+    >>> y = np.array([complex(np.inf, np.nan), np.nan, complex(np.nan, np.inf)])
+    >>> np.nan_to_num(y)
+    array([1.79769313e+308 +0.00000000e+000j, # may vary
+           0.00000000e+000 +0.00000000e+000j,
+           0.00000000e+000 +1.79769313e+308j])
+    >>> np.nan_to_num(y, nan=111111, posinf=222222)
+    array([222222.+111111.j, 111111.     +0.j, 111111.+222222.j])
+
+    """
+
+    dpnp.check_supported_arrays_type(x)
+
+    # Python boolean is a subtype of an integer
+    # so additional check for bool is not needed.
+    if not isinstance(nan, (int, float)):
+        raise TypeError(
+            "nan must be a scalar of an integer, float, bool, "
+            f"but got {type(nan)}"
+        )
+
+    out = dpnp.empty_like(x) if copy else x
+    x_type = x.dtype.type
+
+    if not issubclass(x_type, dpnp.inexact):
+        return x
+
+    parts = (
+        (x.real, x.imag) if issubclass(x_type, dpnp.complexfloating) else (x,)
+    )
+    parts_out = (
+        (out.real, out.imag)
+        if issubclass(x_type, dpnp.complexfloating)
+        else (out,)
+    )
+    max_f, min_f = _get_max_min(x.real.dtype)
+    if posinf is not None:
+        if not isinstance(posinf, (int, float)):
+            raise TypeError(
+                "posinf must be a scalar of an integer, float, bool, "
+                f"or be None, but got {type(posinf)}"
+            )
+        max_f = posinf
+    if neginf is not None:
+        if not isinstance(neginf, (int, float)):
+            raise TypeError(
+                "neginf must be a scalar of an integer, float, bool, "
+                f"or be None, but got {type(neginf)}"
+            )
+        min_f = neginf
+
+    for part, part_out in zip(parts, parts_out):
+        nan_mask = dpnp.isnan(part)
+        posinf_mask = dpnp.isposinf(part)
+        neginf_mask = dpnp.isneginf(part)
+
+        part = dpnp.where(nan_mask, nan, part, out=part_out)
+        part = dpnp.where(posinf_mask, max_f, part, out=part_out)
+        part = dpnp.where(neginf_mask, min_f, part, out=part_out)
+
+    return out
+
+
 _NEGATIVE_DOCSTRING = """
 Computes the numerical negative for each element `x_i` of input array `x`.
 

tests/skipped_tests.tbl

@@ -206,22 +206,6 @@ tests/third_party/cupy/manipulation_tests/test_dims.py::TestInvalidBroadcast_par
 tests/third_party/cupy/manipulation_tests/test_dims.py::TestInvalidBroadcast_param_2_{shapes=[(3, 2), (3, 4)]}::test_invalid_broadcast
 tests/third_party/cupy/manipulation_tests/test_dims.py::TestInvalidBroadcast_param_3_{shapes=[(0,), (2,)]}::test_invalid_broadcast
 
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_negative
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_for_old_numpy
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_negative_for_old_numpy
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inf
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_nan
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inf_nan
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_nan_arg
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inf_arg
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_broadcast[nan]
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_broadcast[posinf]
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_broadcast[neginf]
-
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_scalar_nan
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_copy
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inplace
 tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_real_if_close_real_dtypes
 tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_real_if_close_with_tol_real_dtypes
 tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_real_if_close_true

tests/skipped_tests_gpu.tbl

@@ -260,22 +260,6 @@ tests/third_party/cupy/manipulation_tests/test_dims.py::TestInvalidBroadcast_par
 tests/third_party/cupy/manipulation_tests/test_dims.py::TestInvalidBroadcast_param_2_{shapes=[(3, 2), (3, 4)]}::test_invalid_broadcast
 tests/third_party/cupy/manipulation_tests/test_dims.py::TestInvalidBroadcast_param_3_{shapes=[(0,), (2,)]}::test_invalid_broadcast
 
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_negative
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_for_old_numpy
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_negative_for_old_numpy
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inf
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_nan
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inf_nan
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_nan_arg
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inf_arg
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_broadcast[nan]
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_broadcast[posinf]
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_broadcast[neginf]
-
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_scalar_nan
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_copy
-tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_nan_to_num_inplace
 tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_real_if_close_real_dtypes
 tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_real_if_close_with_tol_real_dtypes
 tests/third_party/cupy/math_tests/test_misc.py::TestMisc::test_real_if_close_true

tests/test_mathematical.py

@@ -1116,6 +1116,66 @@ def test_subtract(self, dtype, lhs, rhs):
         self._test_mathematical("subtract", dtype, lhs, rhs, check_type=False)
 
 
+class TestNanToNum:
+    @pytest.mark.parametrize("dtype", get_all_dtypes(no_none=True))
+    @pytest.mark.parametrize("shape", [(3,), (2, 3), (3, 2, 2)])
+    def test_nan_to_num(self, dtype, shape):
+        a = numpy.random.randn(*shape).astype(dtype)
+        if not dpnp.issubdtype(dtype, dpnp.integer):
+            a.flat[1] = numpy.nan
+        a_dp = dpnp.array(a)
+
+        result = dpnp.nan_to_num(a_dp)
+        expected = numpy.nan_to_num(a)
+        assert_allclose(result, expected)
+
+    @pytest.mark.parametrize(
+        "data", [[], [numpy.nan], [numpy.inf], [-numpy.inf]]
+    )
+    @pytest.mark.parametrize("dtype", get_float_complex_dtypes())
+    def test_empty_and_single_value_arrays(self, data, dtype):
+        a = numpy.array(data, dtype)
+        ia = dpnp.array(a)
+
+        result = dpnp.nan_to_num(ia)
+        expected = numpy.nan_to_num(a)
+        assert_allclose(result, expected)
+
+    def test_boolean_array(self):
+        a = numpy.array([True, False, numpy.nan], dtype=bool)
+        ia = dpnp.array(a)
+
+        result = dpnp.nan_to_num(ia)
+        expected = numpy.nan_to_num(a)
+        assert_allclose(result, expected)
+
+    def test_errors(self):
+        ia = dpnp.array([0, 1, dpnp.nan, dpnp.inf, -dpnp.inf])
+
+        # unsupported type `a`
+        a_np = dpnp.asnumpy(ia)
+        assert_raises(TypeError, dpnp.nan_to_num, a_np)
+
+        # unsupported type `nan`
+        i_nan = dpnp.array(1)
+        assert_raises(TypeError, dpnp.nan_to_num, ia, nan=i_nan)
+
+        # unsupported type `posinf`
+        i_posinf = dpnp.array(1)
+        assert_raises(TypeError, dpnp.nan_to_num, ia, posinf=i_posinf)
+
+        # unsupported type `neginf`
+        i_neginf = dpnp.array(1)
+        assert_raises(TypeError, dpnp.nan_to_num, ia, neginf=i_neginf)
+
+    @pytest.mark.parametrize("kwarg", ["nan", "posinf", "neginf"])
+    @pytest.mark.parametrize("value", [1 - 0j, [1, 2], (1,)])
+    def test_errors_diff_types(self, kwarg, value):
+        ia = dpnp.array([0, 1, dpnp.nan, dpnp.inf, -dpnp.inf])
+        with pytest.raises(TypeError):
+            dpnp.nan_to_num(ia, **{kwarg: value})
+
+
 class TestNextafter:
     @pytest.mark.parametrize("dt", get_float_dtypes())
     @pytest.mark.parametrize(

tests/test_sycl_queue.py

@@ -2336,3 +2336,17 @@ def test_astype(device_x, device_y):
     sycl_queue = dpctl.SyclQueue(device_y)
     y = dpnp.astype(x, dtype="f4", device=sycl_queue)
     assert_sycl_queue_equal(y.sycl_queue, sycl_queue)
+
+
+@pytest.mark.parametrize("copy", [True, False], ids=["True", "False"])
+@pytest.mark.parametrize(
+    "device",
+    valid_devices,
+    ids=[device.filter_string for device in valid_devices],
+)
+def test_nan_to_num(copy, device):
+    a = dpnp.array([-dpnp.nan, -1, 0, 1, dpnp.nan], device=device)
+    result = dpnp.nan_to_num(a, copy=copy)
+
+    assert_sycl_queue_equal(result.sycl_queue, a.sycl_queue)
+    assert copy == (result is not a)

tests/test_usm_type.py

@@ -1361,3 +1361,13 @@ def test_histogram_bin_edges(usm_type_v, usm_type_w):
     assert v.usm_type == usm_type_v
     assert w.usm_type == usm_type_w
     assert edges.usm_type == du.get_coerced_usm_type([usm_type_v, usm_type_w])
+
+
+@pytest.mark.parametrize("copy", [True, False], ids=["True", "False"])
+@pytest.mark.parametrize("usm_type_a", list_of_usm_types, ids=list_of_usm_types)
+def test_nan_to_num(copy, usm_type_a):
+    a = dp.array([-dp.nan, -1, 0, 1, dp.nan], usm_type=usm_type_a)
+    result = dp.nan_to_num(a, copy=copy)
+
+    assert result.usm_type == usm_type_a
+    assert copy == (result is not a)

tests/third_party/cupy/math_tests/test_misc.py

@@ -245,6 +245,7 @@ def test_nan_to_num_inf(self):
     def test_nan_to_num_nan(self):
         self.check_unary_nan("nan_to_num")
 
+    @pytest.mark.skip(reason="Scalar input is not supported")
     @testing.numpy_cupy_allclose(atol=1e-5)
     def test_nan_to_num_scalar_nan(self, xp):
         return xp.nan_to_num(xp.nan)
@@ -260,26 +261,27 @@ def test_nan_to_num_inf_arg(self):
 
     @testing.numpy_cupy_array_equal()
     def test_nan_to_num_copy(self, xp):
-        x = xp.asarray([0, 1, xp.nan, 4], dtype=xp.float64)
+        x = xp.asarray([0, 1, xp.nan, 4], dtype=cupy.default_float_type())
         y = xp.nan_to_num(x, copy=True)
         assert x is not y
         return y
 
     @testing.numpy_cupy_array_equal()
     def test_nan_to_num_inplace(self, xp):
-        x = xp.asarray([0, 1, xp.nan, 4], dtype=xp.float64)
+        x = xp.asarray([0, 1, xp.nan, 4], dtype=cupy.default_float_type())
         y = xp.nan_to_num(x, copy=False)
         assert x is y
         return y
 
+    @pytest.mark.skip(reason="nan, posinf, neginf as array are not supported")
     @pytest.mark.parametrize("kwarg", ["nan", "posinf", "neginf"])
     def test_nan_to_num_broadcast(self, kwarg):
         for xp in (numpy, cupy):
-            x = xp.asarray([0, 1, xp.nan, 4], dtype=xp.float64)
-            y = xp.zeros((2, 4), dtype=xp.float64)
-            with pytest.raises(ValueError):
+            x = xp.asarray([0, 1, xp.nan, 4], dtype=cupy.default_float_type())
+            y = xp.zeros((2, 4), dtype=cupy.default_float_type())
+            with pytest.raises(TypeError):
                 xp.nan_to_num(x, **{kwarg: y})
-            with pytest.raises(ValueError):
+            with pytest.raises(TypeError):
                 xp.nan_to_num(0.0, **{kwarg: y})
 
     @testing.for_all_dtypes(no_bool=True, no_complex=True)