[ENH] refactor metric classes to one class per file

[fkiraly]

Refactors the metrics module to contain private files with one metric class per private module each.

The benefit is not having a gigantic file with all metric classes in it, and following the general pattern that is also used for estimators and other object types.

Also fixes some minor errors with regards to compatibility with 1D numpy arrays.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Refactors the skpro.metrics module by extracting individual metric classes into separate private modules and updating the package exports accordingly, replacing the previous monolithic _classes.py.

Changes:

• Split metric implementations into one-class-per-file private modules (e.g., _pinball.py, _logloss.py, _crps.py)
• Removed the old aggregated skpro/metrics/_classes.py
• Updated skpro/metrics/__init__.py to import metrics from the new modules

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
skpro/metrics/_squared_loss.py	New module containing SquaredDistrLoss extracted from the old _classes.py.
skpro/metrics/_pinball.py	New module containing PinballLoss extracted from the old _classes.py.
skpro/metrics/_logloss_linearlized.py	New module containing LinearizedLogLoss extracted from the old _classes.py (but with spelling issues).
skpro/metrics/_logloss.py	New module containing LogLoss extracted from the old _classes.py.
skpro/metrics/_interval_width.py	New module containing IntervalWidth extracted from the old _classes.py.
skpro/metrics/_empirical_coverage.py	New module containing EmpiricalCoverage extracted from the old _classes.py.
skpro/metrics/_crps.py	New module containing CRPS extracted from the old _classes.py.
skpro/metrics/_constraint_violation.py	New module containing ConstraintViolation extracted from the old _classes.py.
skpro/metrics/_classes.py	Removed the previous single-file container of metric classes.
skpro/metrics/_aucc.py	New module containing AUCalibration extracted from the old _classes.py.
skpro/metrics/init.py	Updated public exports to import from the new per-metric modules.

---

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

[Copilot]

y_true_np is the NumPy array, but the reshape is applied to y_true. If y_true is a pandas object (as allowed by the docstring), y_true.reshape(...) will raise. Reshape y_true_np instead (e.g., y_true_np = y_true_np.reshape(-1, 1)).

Suggested change

	y_true_np = y_true.reshape(-1, 1)
	y_true_np = y_true_np.reshape(-1, 1)

[Copilot]

y_true_np is the array you want to reshape, but the code reshapes y_true. This will fail for pandas inputs. Use y_true_np = y_true_np.reshape(-1, 1).

Suggested change

	y_true_np = y_true.reshape(-1, 1)
	y_true_np = y_true_np.reshape(-1, 1)

[Copilot]

Same reshape issue as other interval metrics: y_true may be a pandas object and not support .reshape. Reshape y_true_np instead.

Suggested change

	y_true_np = y_true.reshape(-1, 1)
	y_true_np = y_true_np.reshape(-1, 1)

[Copilot]

The module name _logloss_linearlized appears to be misspelled (should be _logloss_linearized). This makes the codebase harder to navigate and is easy to propagate into downstream imports. Consider renaming the file/module and updating this import accordingly.

Suggested change

	from skpro.metrics._logloss_linearlized import LinearizedLogLoss
	try:
	from skpro.metrics._logloss_linearized import LinearizedLogLoss
	except ImportError:
	from skpro.metrics._logloss_linearlized import LinearizedLogLoss

[Copilot]

Typo in the class docstring: 'Lineararized' should be 'Linearized'.

Suggested change

	r"""Lineararized logarithmic loss for distributional predictions.
	r"""Linearized logarithmic loss for distributional predictions.

[Copilot]

get_test_params is a @classmethod but uses self as the first parameter. While it technically works, it’s misleading and inconsistent with the other metrics in this PR. Rename the first parameter to cls and consider adding the optional parameter_set=\"default\" argument for interface consistency with other metric classes.

Suggested change

	def get_test_params(self):
	def get_test_params(cls, parameter_set="default"):