This is an umbrella issue to track the implementation of the Array API in Dask. See also the earlier discussion in dask/community#109, and a possible implementation in #8750.

Compatible changes to make dask.array more consistent with numpy

• Add expand_dims to Dask array #8667
• concatenate inconsistency with NumPy for axis=None #8668
• eye inconsistency with NumPy for dtype=None #8669
• full inconsistency with NumPy for dtype=None #8947
• linspace division by zero error #8670
• roll division by zero error #8914
• Add missing NumPy ufuncs: abs, left_shift, right_shift, positive #8915
• Support keepdims in argmax and argmin #8916
• argmin errors on 0-dimensional input #8929
• Add constants to dask.array for consistency with NumPy and array API #8931
• matmul doesn't fail fast for shape mismatches in some cases #8948

Breaking changes to make dask.array more consistent with numpy

• Make eye signature consistent with NumPy #8934

Compatible changes to make dask.array more consistent with numpy.array_api (see Array API Standard Compatibility)

• Add correction keyword argument as a synonym for ddof in std and var
• Add copy keyword argument to asarray