The decision to implement a TypeError broke a significant amount of the sklearnex interfaces for dpctl tensors and has prevented us from properly upgrading to 0.20 support for nearly a half of a year. Instead of raising a TypeError suggesting to use asnumpy simply use asnumpy for the __array__ method.
I expect that the solution in #1964 was a stopgap. Was that the case?
The decision to implement a TypeError broke a significant amount of the sklearnex interfaces for dpctl tensors and has prevented us from properly upgrading to 0.20 support for nearly a half of a year. Instead of raising a TypeError suggesting to use
asnumpysimply use asnumpy for the__array__method.I expect that the solution in #1964 was a stopgap. Was that the case?