ENH: Array API 2024.12 binary ops vs. Python scalars (#131)

Co-authored-by: Lucas Colley <[email protected]>

Author: 2 people

docs/conf.py

@@ -53,6 +53,7 @@
 
 intersphinx_mapping = {
     "python": ("https://docs.python.org/3", None),
+    "array-api": ("https://data-apis.org/array-api/draft", None),
     "jax": ("https://jax.readthedocs.io/en/latest", None),
 }
 

src/array_api_extra/_delegation.py

@@ -52,8 +52,8 @@ def isclose(
 
     Parameters
     ----------
-    a, b : Array
-        Input arrays to compare.
+    a, b : Array | int | float | complex | bool
+        Input objects to compare. At least one must be an array.
     rtol : array_like, optional
         The relative tolerance parameter (see Notes).
     atol : array_like, optional

src/array_api_extra/_lib/_funcs.py

@@ -12,6 +12,7 @@
 from ._at import at
 from ._utils import _compat, _helpers
 from ._utils._compat import array_namespace, is_jax_array
+from ._utils._helpers import asarrays
 from ._utils._typing import Array
 
 __all__ = [
@@ -315,6 +316,7 @@ def isclose(
     xp: ModuleType,
 ) -> Array:  # numpydoc ignore=PR01,RT01
     """See docstring in array_api_extra._delegation."""
+    a, b = asarrays(a, b, xp=xp)
 
     a_inexact = xp.isdtype(a.dtype, ("real floating", "complex floating"))
     b_inexact = xp.isdtype(b.dtype, ("real floating", "complex floating"))
@@ -356,8 +358,8 @@ def kron(a: Array, b: Array, /, *, xp: ModuleType | None = None) -> Array:
 
     Parameters
     ----------
-    a, b : array
-        Input arrays.
+    a, b : Array | int | float | complex
+        Input arrays or scalars. At least one must be an array.
     xp : array_namespace, optional
         The standard-compatible namespace for `a` and `b`. Default: infer.
 
@@ -420,10 +422,10 @@ def kron(a: Array, b: Array, /, *, xp: ModuleType | None = None) -> Array:
     """
     if xp is None:
         xp = array_namespace(a, b)
+    a, b = asarrays(a, b, xp=xp)
 
-    b = xp.asarray(b)
     singletons = (1,) * (b.ndim - a.ndim)
-    a = xp.broadcast_to(xp.asarray(a), singletons + a.shape)
+    a = xp.broadcast_to(a, singletons + a.shape)
 
     nd_b, nd_a = b.ndim, a.ndim
     nd_max = max(nd_b, nd_a)
@@ -583,6 +585,7 @@ def setdiff1d(
     """
     if xp is None:
         xp = array_namespace(x1, x2)
+    x1, x2 = asarrays(x1, x2, xp=xp)
 
     if assume_unique:
         x1 = xp.reshape(x1, (-1,))

src/array_api_extra/_lib/_utils/_helpers.py

@@ -4,8 +4,10 @@
 from __future__ import annotations
 
 from types import ModuleType
+from typing import cast
 
 from . import _compat
+from ._compat import is_array_api_obj, is_numpy_array
 from ._typing import Array
 
 __all__ = ["in1d", "mean"]
@@ -91,3 +93,85 @@ def mean(
         mean_imag = xp.mean(x_imag, axis=axis, keepdims=keepdims)
         return mean_real + (mean_imag * xp.asarray(1j))
     return xp.mean(x, axis=axis, keepdims=keepdims)
+
+
+def is_python_scalar(x: object) -> bool:  # numpydoc ignore=PR01,RT01
+    """Return True if `x` is a Python scalar, False otherwise."""
+    # isinstance(x, float) returns True for np.float64
+    # isinstance(x, complex) returns True for np.complex128
+    return isinstance(x, int | float | complex | bool) and not is_numpy_array(x)
+
+
+def asarrays(
+    a: Array | int | float | complex | bool,
+    b: Array | int | float | complex | bool,
+    xp: ModuleType,
+) -> tuple[Array, Array]:
+    """
+    Ensure both `a` and `b` are arrays.
+
+    If `b` is a python scalar, it is converted to the same dtype as `a`, and vice versa.
+
+    Behavior is not specified when mixing a Python ``float`` and an array with an
+    integer data type; this may give ``float32``, ``float64``, or raise an exception.
+    Behavior is implementation-specific.
+
+    Similarly, behavior is not specified when mixing a Python ``complex`` and an array
+    with a real-valued data type; this may give ``complex64``, ``complex128``, or raise
+    an exception. Behavior is implementation-specific.
+
+    Parameters
+    ----------
+    a, b : Array | int | float | complex | bool
+        Input arrays or scalars. At least one must be an array.
+    xp : ModuleType
+        The standard-compatible namespace for the returned arrays.
+
+    Returns
+    -------
+    Array, Array
+        The input arrays, possibly converted to arrays if they were scalars.
+
+    See Also
+    --------
+    mixing-arrays-with-python-scalars : Array API specification for the behavior.
+    """
+    a_scalar = is_python_scalar(a)
+    b_scalar = is_python_scalar(b)
+    if not a_scalar and not b_scalar:
+        return a, b  # This includes misc. malformed input e.g. str
+
+    swap = False
+    if a_scalar:
+        swap = True
+        b, a = a, b
+
+    if is_array_api_obj(a):
+        # a is an Array API object
+        # b is a int | float | complex | bool
+
+        # pyright doesn't like it if you reuse the same variable name
+        xa = cast(Array, a)
+
+        # https://data-apis.org/array-api/draft/API_specification/type_promotion.html#mixing-arrays-with-python-scalars
+        same_dtype = {
+            bool: "bool",
+            int: ("integral", "real floating", "complex floating"),
+            float: ("real floating", "complex floating"),
+            complex: "complex floating",
+        }
+        kind = same_dtype[type(b)]  # type: ignore[index]
+        if xp.isdtype(xa.dtype, kind):
+            xb = xp.asarray(b, dtype=xa.dtype)
+        else:
+            # Undefined behaviour. Let the function deal with it, if it can.
+            xb = xp.asarray(b)
+
+    else:
+        # Neither a nor b are Array API objects.
+        # Note: we can only reach this point when one explicitly passes
+        # xp=xp to the calling function; otherwise we fail earlier on
+        # array_namespace(a, b).
+        xa, xb = xp.asarray(a), xp.asarray(b)
+
+    return (xb, xa) if swap else (xa, xb)

tests/test_funcs.py

@@ -401,6 +401,24 @@ def test_none_shape_bool(self, xp: ModuleType):
         a = a[a]
         xp_assert_equal(isclose(a, b), xp.asarray([True, False]))
 
+    @pytest.mark.skip_xp_backend(Backend.NUMPY_READONLY, reason="xp=xp")
+    @pytest.mark.skip_xp_backend(Backend.TORCH, reason="Array API 2024.12 support")
+    def test_python_scalar(self, xp: ModuleType):
+        a = xp.asarray([0.0, 0.1], dtype=xp.float32)
+        xp_assert_equal(isclose(a, 0.0), xp.asarray([True, False]))
+        xp_assert_equal(isclose(0.0, a), xp.asarray([True, False]))
+
+        a = xp.asarray([0, 1], dtype=xp.int16)
+        xp_assert_equal(isclose(a, 0), xp.asarray([True, False]))
+        xp_assert_equal(isclose(0, a), xp.asarray([True, False]))
+
+        xp_assert_equal(isclose(0, 0, xp=xp), xp.asarray(True))
+        xp_assert_equal(isclose(0, 1, xp=xp), xp.asarray(False))
+
+    def test_all_python_scalars(self):
+        with pytest.raises(TypeError, match="Unrecognized"):
+            isclose(0, 0)
+
     def test_xp(self, xp: ModuleType):
         a = xp.asarray([0.0, 0.0])
         b = xp.asarray([1e-9, 1e-4])
@@ -413,30 +431,22 @@ def test_basic(self, xp: ModuleType):
         # Using 0-dimensional array
         a = xp.asarray(1)
         b = xp.asarray([[1, 2], [3, 4]])
-        k = xp.asarray([[1, 2], [3, 4]])
-        xp_assert_equal(kron(a, b), k)
-        a = xp.asarray([[1, 2], [3, 4]])
-        b = xp.asarray(1)
-        xp_assert_equal(kron(a, b), k)
+        xp_assert_equal(kron(a, b), b)
+        xp_assert_equal(kron(b, a), b)
 
         # Using 1-dimensional array
         a = xp.asarray([3])
         b = xp.asarray([[1, 2], [3, 4]])
         k = xp.asarray([[3, 6], [9, 12]])
         xp_assert_equal(kron(a, b), k)
-        a = xp.asarray([[1, 2], [3, 4]])
-        b = xp.asarray([3])
-        xp_assert_equal(kron(a, b), k)
+        xp_assert_equal(kron(b, a), k)
 
         # Using 3-dimensional array
         a = xp.asarray([[[1]], [[2]]])
         b = xp.asarray([[1, 2], [3, 4]])
         k = xp.asarray([[[1, 2], [3, 4]], [[2, 4], [6, 8]]])
         xp_assert_equal(kron(a, b), k)
-        a = xp.asarray([[1, 2], [3, 4]])
-        b = xp.asarray([[[1]], [[2]]])
-        k = xp.asarray([[[1, 2], [3, 4]], [[2, 4], [6, 8]]])
-        xp_assert_equal(kron(a, b), k)
+        xp_assert_equal(kron(b, a), k)
 
     def test_kron_smoke(self, xp: ModuleType):
         a = xp.ones((3, 3))
@@ -474,6 +484,18 @@ def test_kron_shape(
         k = kron(a, b)
         assert k.shape == expected_shape
 
+    def test_python_scalar(self, xp: ModuleType):
+        a = 1
+        # Test no dtype promotion to xp.asarray(a); use b.dtype
+        b = xp.asarray([[1, 2], [3, 4]], dtype=xp.int16)
+        xp_assert_equal(kron(a, b), b)
+        xp_assert_equal(kron(b, a), b)
+        xp_assert_equal(kron(1, 1, xp=xp), xp.asarray(1))
+
+    def test_all_python_scalars(self):
+        with pytest.raises(TypeError, match="Unrecognized"):
+            kron(1, 1)
+
     def test_device(self, xp: ModuleType, device: Device):
         x1 = xp.asarray([1, 2, 3], device=device)
         x2 = xp.asarray([4, 5], device=device)
@@ -601,6 +623,28 @@ def test_shapes(
         actual = setdiff1d(x1, x2, assume_unique=assume_unique)
         xp_assert_equal(actual, xp.empty((0,)))
 
+    @pytest.mark.skip_xp_backend(Backend.NUMPY_READONLY, reason="xp=xp")
+    @pytest.mark.parametrize("assume_unique", [True, False])
+    def test_python_scalar(self, xp: ModuleType, assume_unique: bool):
+        # Test no dtype promotion to xp.asarray(x2); use x1.dtype
+        x1 = xp.asarray([3, 1, 2], dtype=xp.int16)
+        x2 = 3
+        actual = setdiff1d(x1, x2, assume_unique=assume_unique)
+        xp_assert_equal(actual, xp.asarray([1, 2], dtype=xp.int16))
+
+        actual = setdiff1d(x2, x1, assume_unique=assume_unique)
+        xp_assert_equal(actual, xp.asarray([], dtype=xp.int16))
+
+        xp_assert_equal(
+            setdiff1d(0, 0, assume_unique=assume_unique, xp=xp),
+            xp.asarray([0])[:0],  # Default int dtype for backend
+        )
+
+    @pytest.mark.parametrize("assume_unique", [True, False])
+    def test_all_python_scalars(self, assume_unique: bool):
+        with pytest.raises(TypeError, match="Unrecognized"):
+            setdiff1d(0, 0, assume_unique=assume_unique)
+
     def test_device(self, xp: ModuleType, device: Device):
         x1 = xp.asarray([3, 8, 20], device=device)
         x2 = xp.asarray([2, 3, 4], device=device)

tests/test_utils.py

@@ -1,11 +1,12 @@
 from types import ModuleType
 
+import numpy as np
 import pytest
 
 from array_api_extra._lib import Backend
 from array_api_extra._lib._testing import xp_assert_equal
 from array_api_extra._lib._utils._compat import device as get_device
-from array_api_extra._lib._utils._helpers import in1d
+from array_api_extra._lib._utils._helpers import asarrays, in1d
 from array_api_extra._lib._utils._typing import Device
 from array_api_extra.testing import lazy_xp_function
 
@@ -45,3 +46,97 @@ def test_xp(self, xp: ModuleType):
         expected = xp.asarray([True, False])
         actual = in1d(x1, x2, xp=xp)
         xp_assert_equal(actual, expected)
+
+
+@pytest.mark.skip_xp_backend(Backend.SPARSE, reason="no isdtype")
+@pytest.mark.parametrize(
+    ("dtype", "b", "defined"),
+    [
+        # Well-defined cases of dtype promotion from Python scalar to Array
+        # bool vs. bool
+        ("bool", True, True),
+        # int vs. xp.*int*, xp.float*, xp.complex*
+        ("int16", 1, True),
+        ("uint8", 1, True),
+        ("float32", 1, True),
+        ("float64", 1, True),
+        ("complex64", 1, True),
+        ("complex128", 1, True),
+        # float vs. xp.float, xp.complex
+        ("float32", 1.0, True),
+        ("float64", 1.0, True),
+        ("complex64", 1.0, True),
+        ("complex128", 1.0, True),
+        # complex vs. xp.complex
+        ("complex64", 1.0j, True),
+        ("complex128", 1.0j, True),
+        # Undefined cases
+        ("bool", 1, False),
+        ("int64", 1.0, False),
+        ("float64", 1.0j, False),
+    ],
+)
+def test_asarrays_array_vs_scalar(
+    dtype: str, b: int | float | complex, defined: bool, xp: ModuleType
+):
+    a = xp.asarray(1, dtype=getattr(xp, dtype))
+
+    xa, xb = asarrays(a, b, xp)
+    assert xa.dtype == a.dtype
+    if defined:
+        assert xb.dtype == a.dtype
+    else:
+        assert xb.dtype == xp.asarray(b).dtype
+
+    xbr, xar = asarrays(b, a, xp)
+    assert xar.dtype == xa.dtype
+    assert xbr.dtype == xb.dtype
+
+
+def test_asarrays_scalar_vs_scalar(xp: ModuleType):
+    a, b = asarrays(1, 2.2, xp=xp)
+    assert a.dtype == xp.asarray(1).dtype  # Default dtype
+    assert b.dtype == xp.asarray(2.2).dtype  # Default dtype; not broadcasted
+
+
+ALL_TYPES = (
+    "int8",
+    "int16",
+    "int32",
+    "int64",
+    "uint8",
+    "uint16",
+    "uint32",
+    "uint64",
+    "float32",
+    "float64",
+    "complex64",
+    "complex128",
+    "bool",
+)
+
+
+@pytest.mark.parametrize("a_type", ALL_TYPES)
+@pytest.mark.parametrize("b_type", ALL_TYPES)
+def test_asarrays_array_vs_array(a_type: str, b_type: str, xp: ModuleType):
+    """
+    Test that when both inputs of asarray are already Array API objects,
+    they are returned unchanged.
+    """
+    a = xp.asarray(1, dtype=getattr(xp, a_type))
+    b = xp.asarray(1, dtype=getattr(xp, b_type))
+    xa, xb = asarrays(a, b, xp)
+    assert xa.dtype == a.dtype
+    assert xb.dtype == b.dtype
+
+
+@pytest.mark.parametrize("dtype", [np.float64, np.complex128])
+def test_asarrays_numpy_generics(dtype: type):
+    """
+    Test special case of np.float64 and np.complex128,
+    which are subclasses of float and complex.
+    """
+    a = dtype(0)
+    xa, xb = asarrays(a, 0, xp=np)
+    assert xa.dtype == dtype
+    assert xb.dtype == dtype