Goodbye pytest warnings

[AdrianSosic]

Say goodbye to all pytest warnings, once and for all. This is achieved but turning the warnings into errors using the -W error mode, requiring us to fix the causing issues. Generally good practice since:

• it finds more issues in our own code base (see changes)
• and also prevents us from running e.g. into deprecated features from third-party code (see changes)

So the PR essentially finalizes the work from #766. In addition, it enables the --strict mode.

[Copilot]

Pull request overview

This PR tightens the test suite’s warning hygiene by turning warnings into errors and enabling stricter pytest behavior, then adjusting code/tests to eliminate or suppress the resulting warnings (including a few third‑party deprecations).

Changes:

• Enable stricter pytest configuration/marker handling and convert warnings to errors via pytest.ini (filterwarnings = error plus targeted ignores).
• Update tests to avoid/contain deprecation warnings that occur at collection time (module import / parametrization).
• Adjust runtime code paths to avoid warning-triggering tensor conversions and pandas FutureWarnings.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
pytest.ini	Enables strict pytest modes and turns warnings into errors with targeted ignores.
tests/conftest.py	Adds a temporary global warning ignore and fixes/skips warning-triggering test helpers.
tests/test_deprecations.py	Suppresses deprecations that fire at collection time and in specific tests under -W error behavior.
tests/test_surrogate.py	Adjusts test input size to avoid warning-triggering behavior.
tests/hypothesis_strategies/alternative_creation/test_acquisition.py	Deduplicates acquisition function strings used for parametrization.
tests/constraints/test_constraints_continuous.py	Fixes a duplicated parametrization id.
tests/constraints/test_cardinality_constraint_continuous.py	Ensures the expected warning is always captured under warning-as-error mode.
baybe/constraints/discrete.py	Suppresses pandas FutureWarning during dummy assignments for dependency constraints.
baybe/surrogates/random_forest.py	Avoids converting tensors requiring grad to numpy directly by detaching first.
baybe/surrogates/ngboost.py	Detaches and converts torch tensors before passing to NGBoost prediction API.
baybe/surrogates/linear.py	Detaches and converts torch tensors before passing to sklearn prediction API.
baybe/utils/clustering_algorithms/third_party/kmedoids.py	Attempts to replace deprecated/removed sklearn utility usage in k-medoids++ init.

---

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

[Copilot]

np.cumulative_sum is not a NumPy API, so this will raise AttributeError at runtime. If the goal is to remove the scikit-learn stable_cumsum import, replace this with a valid cumulative sum implementation (e.g., np.cumsum) or inline a numerically-stable cumulative sum equivalent to stable_cumsum to preserve behavior.

Suggested change

	np.cumulative_sum(closest_dist_sq), rand_vals
	np.cumsum(closest_dist_sq), rand_vals

[Copilot]

Using a set for pytest.mark.parametrize makes test collection order non-deterministic (hash-randomized), which can lead to flaky ordering/IDs across runs and platforms. Consider keeping the de-duplication but switching to a deterministic sequence (e.g., sorted(set(...))) before passing it to parametrize.

Suggested change

	combined = set(abbreviations + fullnames)
	combined = sorted(set(abbreviations + fullnames))

[Copilot]

The warnings.simplefilter("ignore", FutureWarning) inside get_invalid suppresses all FutureWarnings raised in this block, which can mask unrelated warnings and make it harder to spot new pandas deprecations. Prefer narrowing the filter to the specific pandas warning message (or module) you're targeting, or proactively adjust the dataframe/column dtypes before assignment to avoid emitting the warning in the first place.

Suggested change

	with warnings.catch_warnings():
	# Suppress pandas FutureWarning about setting incompatible dtype.
	# TODO: Needs proper fix for pandas 3.0 compatibility.
	warnings.simplefilter("ignore", FutureWarning)
	for k, _ in enumerate(self.parameters):
	# .loc assignments are not supported by mypy + pandas-stubs yet
	# See https://github.com/pandas-dev/pandas-stubs/issues/572
	censored_data.loc[ # type: ignore[call-overload]
	~self.conditions[k].evaluate(data[self.parameters[k]]),
	self.affected_parameters[k],
	] = Dummy()
	# Cast affected columns to object before assigning Dummy() values to avoid
	# pandas FutureWarning about setting incompatible dtypes.
	all_affected_params = list(
	dict.fromkeys(
	affected_param
	for affected_params in self.affected_parameters
	for affected_param in affected_params
	)
	)
	if all_affected_params:
	censored_data[all_affected_params] = censored_data[
	all_affected_params
	].astype(object)
	for k, _ in enumerate(self.parameters):
	# .loc assignments are not supported by mypy + pandas-stubs yet
	# See https://github.com/pandas-dev/pandas-stubs/issues/572
	censored_data.loc[ # type: ignore[call-overload]
	~self.conditions[k].evaluate(data[self.parameters[k]]),
	self.affected_parameters[k],
	] = Dummy()

[Copilot]

The filterwarnings rule on this line has trailing whitespace after the .*botorch/fit regex. Depending on how the config is parsed, this can accidentally become part of the regex and prevent the ignore from matching. Please remove the trailing spaces to make the rule unambiguous.

Suggested change

	ignore:`scipy_minimize` terminated with status .*:botorch.exceptions.warnings.OptimizationWarning:.*botorch/fit
	ignore:`scipy_minimize` terminated with status .*:botorch.exceptions.warnings.OptimizationWarning:.*botorch/fit