ENH TransformedTargetRegressor.fit() raises if only inverse_func is provided

[StefanieSenger]

What does this implement/fix? Explain your changes.

This PR suggests to raise an error if users only provide inverse_func to TransformedTargetRegressor() without explicitely setting func as well.

It's losely connected to #28480, where I came across this issue.

Not sure if I should add a changelog entry here.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: fd3fd7c. Link to the linter CI: here}

[jeremiedbb]

Thanks for the PR @StefanieSenger.

Currently when inverse_func is passed and func is left to None, func falls back to identity and we get a warning because both functions are not their respective inverse.

I agree that falling back to identity when inverse_func is passed is not a desirable behavior so I'm +1 to raise an error. Since it's a change of behavior, it requires an entry in the change log though.

[jeremiedbb]

You can simplify this condition but I'm not sure it helps readability so feel free to ignore :)

Suggested change

	if (self.func is not None and self.inverse_func is None) or (
	self.inverse_func is not None and self.func is None
	):
	if (self.func is not None) == (self.inverse_func is None):

[StefanieSenger]

I think that this is difficult to read and would prefer to keep the other code, if that's okay.

[StefanieSenger]

Thanks @jeremiedbb, I should have explained it the way you did.
I also think that in this case we should force the user to input something explicitly. The warning then is for other inputs where func and inverse_func don't match.

[thomasjpfan]

Two nits, otherwise LGTM

[jeremiedbb]

Please also add a what's new entry

[StefanieSenger]

I've applied your suggestions, @thomasjpfan and added the changelog entry, @jeremiedbb.
Thank you both for reviewing.