We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
dpt.take_along_axis
dpt.put_along_axis
uint64
In dpctl, take_along_axis raises an error when indices dtype is uint64 while it works properly in NumPy.
dpctl
take_along_axis
import dpctl.tensor as dpt a = dpt.asarray([[10, 30, 20], [60, 40, 50]]) ind = dpt.asarray([[2, 1, 0], [2, 1, 0]], dtype=dpt.uint64) dpt.take_along_axis(a, ind, axis=1) # ValueError: cannot safely promote indices to an integer data type import numpy a = numpy.array([[10, 30, 20], [60, 40, 50]]) ind = numpy.array([[2, 1, 0], [2, 1, 0]], dtype=numpy.uint64) numpy.take_along_axis(a, ind, axis=1) # array([[20, 30, 10], # [50, 40, 60]])
a similar behavior is observed for dpt.put_along_axis