TSC refactor

Time series classifiers should be refactored according to the new extension template.

If you would like to work on this: please post here and which classifier you are taking on, to avoid duplication of work. A core dev will then add your name to the classifier under refactor and/or link the PR.

Checklist of classifiers:

• BaseColumnEnsembleClassifier
• ComposableTimeSeriesForestClassifier
• BOSSEnsemble, IndividualBOSS https://github.com/alan-turing-institute/sktime/pull/1544
• ContractableBOSS https://github.com/alan-turing-institute/sktime/pull/1295
• MUSE https://github.com/alan-turing-institute/sktime/pull/1359
• TemporalDictionaryEnsemble, IndividualTDE https://github.com/alan-turing-institute/sktime/pull/1357
• WEASEL https://github.com/alan-turing-institute/sktime/pull/1544
• MUSE https://github.com/alan-turing-institute/sktime/pull/1544
• ElasticEnsemble #1584
• ProximityForest, ProximityStump, ProximityTree #1584
• ShapeDTW https://github.com/alan-turing-institute/sktime/pull/1554
• KNeighborsTimeSeriesClassifier (inherits from _KNeighborsClassifier and calls a version of self._fit in its current fit)
• Catch22Classifier #1487
• MatrixProfileClassifier https://github.com/alan-turing-institute/sktime/pull/1545
• SignatureClassifier https://github.com/alan-turing-institute/sktime/pull/1545
• TSFreshClassifier #1473
• HIVECOTEV1 #1504
• CanonicalIntervalForest https://github.com/alan-turing-institute/sktime/pull/1269
• DrCIF https://github.com/alan-turing-institute/sktime/pull/1269
• TimeSeriesForestClassifier (bit complicated, involves the series_as_features base class)
• RandomIntervalSpectralForest #1583
• SupervisedTimeSeriesForest #1583
• Arsenal https://github.com/alan-turing-institute/sktime/pull/1305
• ROCKETClassifier https://github.com/alan-turing-institute/sktime/pull/1239
• MrSEQL (deprecated)
• ShapeletTransformClassifier #1490

A general recipe for the refactor is as follows:

• go through all the public methods, fit, predict, etc, and change them to the new private version _fit, _predict, etc
• remove input checks (e.g., check_X) and output operations (e.g., set cutoff) from the private version; make sure these are already all present in the BaseClassifier.
• if there’s an input or output check/operation that looks like it should be in BaseClassifier but isn’t, add a note to your PR and make it very explicit; consider adding it to BaseClassifier but make sure @TonyBagnall and @fkiraly are aware of such a change (ping in PR and ask for review).
• make sure all the tests pass. Don’t change any of the tests. (unless there is a really good reason, in which case please explain)
• if there’s an inheritance tree (not just a single class) downstream from BaseClassifier, make sure you do the above for all nodes in the inheritance tree
• move any “properties” or “capabilities” fields to the _tags dictionary.
• If new tags are added, ensure they are added to the tag registry. Look up the classifier tags (using registry.all_tags) before you add any new ones. Ensure @TonyBagnall and @fkiraly are aware of changes to the tag registry (ping in PR and ask for review).

[TonyBagnall]

Hi, sure no problem, can you do it relatively soon? @MatthewMiddlehurst can you leave distance based for now?

[MatthewMiddlehurst]

Both the hybrid Catch22ForestClassifier and the signature_based package were deprecated and moved to feature_based already. This was done a while back, should be fine to remove the old files now.