Jax devicearray prod

[VictorOdede]

Close #19417

[fnhirwa]

Hey @VictorOdede

Thanks for the PR, all looks great just a few changes are needed

[fnhirwa]

I can't fully understand why this is the only supported dtype

[VictorOdede]

Hey @hirwa-nshuti , thanks for the review. I've also tried testing with other valid dtypes but the ground-truth is always returning int64 causing the tests to fail with an AssertionError since returned dtype is different from ground truth. I haven't encountered this problem before, is there a way to solve this issue?

[fnhirwa]

Then if there is a special promotion being done should be handled in the frontend function not on the testing side to ensure that we cover all the corner cases

[VictorOdede]

Hey @hirwa-nshuti , I've added type promotion for all int and uint dtypes. Now all tests are passing except for Paddle which sometimes fails because it doesn't support uint64

[fnhirwa]

Hey
Just a few suggested changes🙂
Keep up the good work😊

[fnhirwa]

From my understandings is that the promotion being done here should be done on the jax.numpy.product function as we can promote all unsigned integers to uint64 by a simple ivy.is_uint_dtype and ivy.is_int_dtype functions.

Here we will simply need to call the frontend function once we ensured that it is working correctly🙂

[VictorOdede]

Thanks for the great pointers. Will look into this asap.

[VictorOdede]

Hey @hirwa-nshuti , the tests are passing for the most part but sometimes it fails for higher precision dtypes like this:

Is this the expected behaviour?

[fnhirwa]

For precision errors handling you can use the torelance values mentioned in the helper function

https://github.com/unifyai/ivy/blob/356e6e5654970ee4b6f7cb9b41b3b572436e1262/ivy_tests/test_ivy/helpers/function_testing.py#L485C4-L485C4

[VictorOdede]

For precision errors handling you can use the torelance values mentioned in the helper function

https://github.com/unifyai/ivy/blob/356e6e5654970ee4b6f7cb9b41b3b572436e1262/ivy_tests/test_ivy/helpers/function_testing.py#L485C4-L485C4

Thanks! Will look into this.

[fnhirwa]

Hey,
Great work so far I have added the suggestions in the comments below
Thanks

[fnhirwa]

These are only supported when the promote_integers is True

[fnhirwa]

We can redefine prod function as

@to_ivy_arrays_and_back
def product(
    a,
    *,
    axis=None,
    dtype=None,
    keepdims=False,
    initial=None,
    where=None,
    promote_integers=True,
    out=None,
):
    if ivy.is_array(where):
        a = ivy.where(where, a, ivy.default(out, ivy.ones_like(a)), out=out)
    if promote_integers:
        if ivy.is_uint_dtype(a.dtype):
            dtype = "uint64"
        elif ivy.is_int_dtype(a.dtype):
            dtype = "int64"
    if initial is not None:
        if axis is not None:
            s = ivy.to_list(ivy.shape(a, as_array=True))
            s[axis] = 1
            header = ivy.full(ivy.Shape(tuple(s)), initial)
            a = ivy.concat([header, a], axis=axis)
        else:
            a[0] *= initial
    return ivy.prod(a, axis=axis, dtype=dtype, keepdims=keepdims, out=out)
    ```

[VictorOdede]

Hey, Great work so far I have added the suggestions in the comments below Thanks

Thanks, will work on implementing this

[fnhirwa]

LGTM!
Going to merge