BUG: isclose: scalars cannot be input alongside non-default device arrays

[OmarManzoor]

When we use isclose with array-api-strict while comparing an array on a device to a scalar value we get the error

ValueError: Arrays from two different devices (array_api_strict.Device('device1') and array_api_strict.Device('CPU_DEVICE')) can not be combined.

Here is a simple example to reproduce this:

import array_api_strict as xp
from array_api_extra import isclose

arr = xp.asarray([0.1, 0.2, 0.3, 0.4, 0.5], device=xp.Device("device1"))
isclose(arr, 1.0)

Can this be adjusted in the array_api_extra implementation to handle the necessary conversions so that we don't have to do them ourselves?

[lucascolley]

This is at least intentional: https://github.com/data-apis/array-api-extra/blob/main/docs/index.md#L144-L146

[lucascolley]

Would you like to make an argument for isclose being a special case, or changing that policy wholesale?

[OmarManzoor]

Lets ask @ogrisel for his opinion

[ogrisel]

For convenience reasons, it might be interesting to also accept Python scalar inputs for functions that accept multiple arrays.

This would be consistent with the spirit of the specification for functions such as xp.where which accepts both arrays an Python scalars:

https://data-apis.org/array-api/latest/API_specification/generated/array_api.where.html

Notes

At least one of x1 and x2 must be an array.

If either x1 or x2 is a scalar value, the returned array must have a data type determined according to Mixing arrays with Python scalars.

[lucascolley]

sounds reasonable to me! Would someone be willing to work on a PR so we can see what changes would be needed?

[OmarManzoor]

I can try to work on it