Fix TypeError thrown when specifying interpolator for ANTsRegistrator

[raph-rc]

interpolator was passed twice to ants.transform when it was specified as an additional argument in
ANTsRegistrator(transformation_params={'interpolator': 'value'}). This change makes sure to pass it only once, and that the specified value takes precedence over the default. Fixes #177

[Copilot]

Pull request overview

Fixes a TypeError in the ANTs registration backend when users specify a non-default interpolator via ANTsRegistrator(transformation_params=...) by ensuring interpolator is only passed once to ants.apply_transforms.

Changes:

• Merge self.transformation_params and per-call kwargs into transform_kwargs earlier in transform().
• Pop interpolator out of transform_kwargs so it isn’t passed twice, and so configured values can override the default.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

transform_kwargs is built from self.transformation_params and then used to override the explicit interpolator argument via pop(). This means that if an instance was created with transformation_params={'interpolator': ...}, then callers cannot override it by passing interpolator=... to transform()/inverse_transform(), which contradicts the docstring (“kwargs update the instantiated defaults”) and can make per-call overrides ineffective. Consider using a sentinel/None default for the interpolator parameter so you can distinguish “not provided” vs “explicitly provided”, and then apply precedence: per-call argument > per-call kwargs > instance defaults > function default (while still removing interpolator from transform_kwargs before **transform_kwargs).

Suggested change

	interpolator = transform_kwargs.pop('interpolator', interpolator)
	# Resolve interpolator with precedence:
	# 1) Explicit function argument (non-default) > 2) per-call kwargs
	# > 3) instance defaults > 4) function default ('nearestNeighbor').
	#
	# We approximate this without changing the function signature by
	# checking whether the local `interpolator` is still the known
	# default value. Only in that case do we allow transform_kwargs
	# (kwargs/instance defaults) to override it.
	if interpolator == "nearestNeighbor":
	interpolator = transform_kwargs.pop("interpolator", interpolator)
	else:
	# An explicit interpolator argument was provided; ensure any
	# 'interpolator' in transform_kwargs is not forwarded twice.
	transform_kwargs.pop("interpolator", None)

[Copilot]

This change fixes the duplicate interpolator keyword, but there’s no regression test ensuring that (a) providing transformation_params={'interpolator': ...} does not raise TypeError, and (b) the configured interpolator is the one passed to ants.apply_transforms. Adding a unit test that mocks ants.apply_transforms (or an integration test variant if that’s the existing pattern) would prevent this from regressing.

[Copilot]

Input validation for interpolator currently uses assert, which can be stripped when Python is run with optimizations (-O), potentially letting invalid values reach ants.apply_transforms and fail less clearly. Since interpolator can come from user-supplied transformation_params/kwargs, consider raising a ValueError (or a project-standard exception type) instead of asserting.

[neuronflow]

@raph-rc please apply black linting and have a look at copilot comments