[ENH] Added R-Clustering clusterer to aeon

[Ramana-Raja]

Reference Issues/PRs

#2132

What does this implement/fix? Explain your changes.

added R clustering model for aeon

Does your contribution introduce a new dependency? If yes, which one?

no

Any other comments?

PR checklist

For all contributions

• I've added myself to the list of contributors. Alternatively, you can use the @all-contributors bot to do this for you.
• The PR title starts with either [ENH], [MNT], [DOC], [BUG], [REF], [DEP] or [GOV] indicating whether the PR topic is related to enhancement, maintenance, documentation, bugs, refactoring, deprecation or governance.

For new estimators and functions

• I've added the estimator to the online API documentation.
• (OPTIONAL) I've added myself as a __maintainer__ at the top of relevant files and want to be contacted regarding its maintenance. Unmaintained files may be removed. This is for the full file, and you should not add yourself if you are just making minor changes or do not want to help maintain its contents.

For developers with write access

• (OPTIONAL) I've updated aeon's CODEOWNERS to receive notifications about future changes to these files.

[aeon-actions-bot]

Thank you for contributing to aeon

I have added the following labels to this PR based on the title: [ $\color{#FEF1BE}{\textsf{enhancement}}$ ].
I have added the following labels to this PR based on the changes made: [ $\color{#4011F3}{\textsf{clustering}}$ ]. Feel free to change these if they do not properly represent the PR.

The Checks tab will show the status of our automated tests. You can click on individual test runs in the tab or "Details" in the panel below to see more information if there is a failure.

If our pre-commit code quality check fails, any trivial fixes will automatically be pushed to your PR unless it is a draft.

Don't hesitate to ask questions on the aeon Slack channel if you have any.

PR CI actions

These checkboxes will add labels to enable/disable CI functionality for this PR. This may not take effect immediately, and a new commit may be required to run the new configuration.

• Run pre-commit checks for all files
• Run mypy typecheck tests
• Run all pytest tests and configurations
• Run all notebook example tests
• Run numba-disabled codecov tests
• Stop automatic pre-commit fixes (always disabled for drafts)
• Disable numba cache loading
• Push an empty commit to re-run CI checks

[TonyBagnall]

hi, thanks for this but if we include this clusterer we want it to use our version of Rocket transformers which are optimised for numba

[Ramana-Raja]

hi, thanks for this but if we include this clusterer we want it to use our version of Rocket transformers which are optimised for numba

sure, I will try to reimplement it and use aeon Rocket transformers

[Ramana-Raja]

Hi @chrisholder, I discovered that the issue was caused by how self._random_state = check_random_state(self.random_state) was being handled. Removing it resolved the problem, though I'm not entirely sure why. Perhaps @MatthewMiddlehurst could provide some insights. Additionally, I removed the estimator as an input for the class, as the research paper only utilized KMeans

[chrisholder]

Looks good - Final thing before approving have you checked equivalence to the original. It's doesnt have to be completely 100% the same (due to random state). However, we should check for 10 or so datasets (which you can load from aeon.datasets) that the scores are similar.

This doesn't need to be an actual test in aeon but if you could run it locally and post the adjusted rand index using sklearn (https://scikit-learn.org/stable/modules/generated/sklearn.metrics.adjusted_rand_score.html) the we can check this version matches the original!

[Ramana-Raja]

Looks good - Final thing before approving have you checked equivalence to the original. It's doesnt have to be completely 100% the same (due to random state). However, we should check for 10 or so datasets (which you can load from aeon.datasets) that the scores are similar.

This doesn't need to be an actual test in aeon but if you could run it locally and post the adjusted rand index using sklearn (https://scikit-learn.org/stable/modules/generated/sklearn.metrics.adjusted_rand_score.html) the we can check this version matches the original!

I tested the ARI scores between the this RCluster and the original RCluster on the experimental dataset provided in the original RCluster code. Here are the results:

[MatthewMiddlehurst]

Not ready sorry.

Please add some code comments explaining the use of private functions from another estimator and the multiple PCAs if you can. Don't usually request it but it is pretty confusing as is.

Class attributes set in fit and fit_predict (self.) should start with _ (protected/private) or end with _ (public and should be documented in the docstring under Attributes)

[MatthewMiddlehurst]

don't set these here

[Ramana-Raja]

@chrisholder While reviewing the code, I identified an architectural issue. The R cluster implementation cannot have separate fit and predict methods because it relies on PCA for dimensionality reduction. For example, if PCA determines the optimal dimension during training to be 13, and we attempt to predict on test data with fewer than 13 dimensions, it results in an error. Even if we create a new PCA and apply fit_transform on the test data, we would need to retrain KMeans, which is what I did. However, I don't believe this is an optimal solution. Could you suggest a better approach to handle this scenario?

[TonyBagnall]

@chrisholder While reviewing the code, I identified an architectural issue. The R cluster implementation cannot have separate fit and predict methods because it relies on PCA for dimensionality reduction. For example, if PCA determines the optimal dimension during training to be 13, and we attempt to predict on test data with fewer than 13 dimensions, it results in an error. Even if we create a new PCA and apply fit_transform on the test data, we would need to retrain KMeans, which is what I did. However, I don't believe this is an optimal solution. Could you suggest a better approach to handle this scenario?

PCA has a separate fit and transform step, so you can PCA fit_transform in fit, save the transform then just PCA transform in transform?

[Ramana-Raja]

@chrisholder While reviewing the code, I identified an architectural issue. The R cluster implementation cannot have separate fit and predict methods because it relies on PCA for dimensionality reduction. For example, if PCA determines the optimal dimension during training to be 13, and we attempt to predict on test data with fewer than 13 dimensions, it results in an error. Even if we create a new PCA and apply fit_transform on the test data, we would need to retrain KMeans, which is what I did. However, I don't believe this is an optimal solution. Could you suggest a better approach to handle this scenario?

PCA has a separate fit and transform step, so you can PCA fit_transform in fit, save the transform then just PCA transform in transform?

If we transform test data(while predicting that is predicting this test data without fitting it using _predict method) with a number of features less than the n_components of PCA, it will cause an error, So we cannot predict test data having number of features less than that of n_components of PCA.