From c94f70e08f56ef7142f8db9f6dc467a28fbf4c2a Mon Sep 17 00:00:00 2001
From: asmeurer Remove the Use a more robust implementation of Fix the definition of Correctly use the array’s device when promoting scalars. (Thanks to
+@betatim) Correctly propagate the input array’s device in Changelog¶
2.1 (2024-10-18)¶
-2.1.1 (2024-11-07)¶
Major Changes¶
+Major Changes¶
+
+
+__array__
method from array-api-strict arrays. This means they
+will no longer be implicitly converted to NumPy arrays when passed to np
+functions. This method was previously implemented as a convenience, but it
+isn’t part of the array API standard. To portably convert an array API
+strict array to a NumPy array, use np.from_dlpack(x)
Minor Changes¶
+
+
+clip()
that handles corner cases better.sign()
for complex numbers when using NumPy 1.x.asarray()
. (Thanks to
+@betatim)2.1 (2024-10-18)¶
+Major Changes¶
Avoid implicitly relying on __array__
in some places. These changes should
not be usef visible.
array-api-strict has a new set of flags that can be used to dynamically enable or disable features in array-api-strict. These @@ -306,23 +330,23 @@
Calling iter()
on an array now correctly raises TypeError
.
Add some missing names to __all__
.
Fix the api_version
argument to __array_namespace__
to accept
'2021.12'
or '2022.12'
.
Fix the copy
flag in __array__
for NumPy 2.0.
Add full copy=False
support to asarray()
. This is emulated in NumPy 1.26 by creating
@@ -331,8 +355,8 @@
This is the first release of array_api_strict
. It is extracted from
numpy.array_api
, which was included as an experimental submodule in NumPy
versions prior to 2.0. Note that the commit history in this repository is
@@ -416,23 +440,28 @@