[ENH] kNN Classifier and Regressor reimplementations

[GuiArcencio]

Reference Issues/PRs

• Fixes [ENH] Redesign KNN classifier and regressor #39.

What does this implement/fix? Explain your changes.

• Reimplemented both KNeighborsTimeSeriesClassifier and KNeighborsTimeSeriesRegressor in order to fix memory leaks and replace hard coded "distances" string list.
• Removed computation of distance matrices in fit (which was useless in any case) and predict, as well as sklearn's KNeighbors instances contained within the models. The k-Neighbors algorithm is now implemented here.
• Nearest neighbors are found by np.argpartitioning the distance vector into [0..k-1], [k], [k+1..], which is $O(n)$ as opposed to the $O(n \log n)$ in sorting.
• Distance metrics are now generated by distance_factory, uncoupling the selection of possible metrics from the model.

What should a reviewer concentrate their feedback on?

• Documentation for the new implementations might be lacking or inconsistent.
• Other mtypes could be added for unequal-length support.
• n_jobs is a parameter for the new classifier for compatibility purposes only. There is no parallelism implementation yet.

The following graphs show the results of current and new implementation benchmarks. The experiments were made on a regression dataset which consists of univariate time series of length 365. At each sample size, data was split between train and test in 70% - 30% proportions, respectively. The distance was always set to 'euclidean'.

More benchmarks are underway, one using distance='dtw' and two others fixing train size and varying test size, and vice-versa.

Did you add any tests for the change?

Some tests were removed and/or changed due to being implementation-specific.

PR checklist

For all contributions

• I've added myself to the list of contributors.
• Optionally, I've updated sktime's CODEOWNERS to receive notifications about future changes to these files.
• The PR title starts with either [ENH], [MNT], [DOC], or [BUG] indicating whether the PR topic is related to enhancement, maintenance, documentation, or bug.

[TonyBagnall]

fantastic, thanks @GuiArcencio. @chrisholder could you take a look please?

[TonyBagnall]

hey @GuiArcencio do you have those timing/memory graphs? If so, good to post them here

[TonyBagnall]

looks really good, thanks!

[TonyBagnall]

looks good to me