sktime/skpro

In sktime, it is possible to specify modules that contain tests for a given estimator that should run in the VM-based test suite, if tests:vm is specified. This has not been added in skpro, we should upgrade the status of the skpro test and tag system to the state of sktime. For testing, we should add tests:specific tags populated at least for some of the estimators.

We should ensure that all scikit-base classes in skpro have at least one usage example. The recipe to find all those without: from sktime.registry import all_estimators [x for x in all_estimators() if ">>>" not in x.__doc__] Once the docstring is added, the _tags should be checked for the line "tests:skip_by_name": ["test_class_has_doctest_example"], which skips the test requiring a docstring. If present, this line should also be removed. Please post here which estimator/object you are working on before starting. If you find a bug with the estimator, e.g., unexpected failures, in the process of adding an example, please open a new issue. The recipe for adding examples is: add an example section in numpydoc stype use as template for a simple example: MultipleQuantileRegressor use as template for complex/multiple examples: NaiveProbaRegressor (search in code base) To pick this up, post an estimator here that you are working on, to avoid double work, and make a PR.

We should add, where known, explicit formulae to the docstrings of distribution methods. This is done by populating the _formula_docs attribute of the given distribution, see the extension template in extension_templates.distributions for a detail explanation, and Normal for an example of a populated distribution. Contributors can help by: adding formulae to a distribution which does not have them yet adding formulae for more methods in a distribution which already has them Please post here before starting, to avoid duplication: which distribution and method(s) you are working on. If opening a PR, please provide a reference for the formula - either a literature reference, or your own compitation. No AI generated formulae please, since these tend to be hallucinated. Recipe for finding distributions that have not yet been dealt with: from skpro.registry import all_objects all_dists = all_objects("distribution", return_names=False) dists_without_formula = [d for d in all_dists if not h

hyperactive has a tuner that can be used for skpro tuning with arbitrary hyperparameter optimization backend. However, it is not visible from the skpro documentation. This should be added using the same forwarding pattern as sktime Prophetverse. It may be that it is necessary to add the _placeholder_record to skpro for this.

Even though scikit-learn supports probabilistic classification and, thus, discrete output distributions, it has a few problems: the predict_proba API is inconsistent with the skpro regression API, as it returns a numpy array and not a distribution object there is no API for multivariate probabilistic classification, or ordinal regression. It was suggested by @felipeangelimvieira to add a skpro native probabilistic classifier API that can represent a wider range of return distributions. Someone working on this should give more details about the API design first. As a starting point, I suggest: a new module classification a base class BaseProbaClassifier with methods fit, predict and predict_proba if ordinal classification is also covered, then predict_quantiles and predict_interval may also make sense, in this case a capability tag for ordinal classification a distribution DiscreteClass used for the output a scikit-learn adapter that allows to expose any sklearn probabilistic classif