Fix

[emekaokoli19]

Description

This PR updates the Unique operator to correctly follow the new NumPy 2.0 semantics for the return_inverse output.

NumPy 2.0 changed the shape of the inverse array:

When axis=None: inverse has the same shape as the input.

When axis=i: inverse has shape (input.shape[i],) instead of always being 1D and flattened.

PyTensor’s infer_shape() and perform() were updated for NumPy 2.0 behavior, but make_node() still declared the return_inverse output as a fixed 1D tensor shape=(None,), causing shape inference inconsistencies and failing tests.

This PR fixes the bug by updating make_node() to construct the return_inverse output with the correct shape based on NumPy 2.0 rules. After this change, the behaviour of make_node, infer_shape, and NumPy’s runtime is fully aligned, and all Unique shape inference tests pass.

Related Issue

• Closes Switch Unique behavior from numpy < 2.0 to numpy >= 2.0 #1739
• Related to #

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):

[emekaokoli19]

@ricardoV94 whenever free please take a look