Tslearn

Added

* Support of the `PyTorch` backend for the metrics of `tslearn`.
In particular, the Dynamic Time Warping (DTW) metric and the Soft-DTW metric now support the `PyTorch` backend.

Removed

* Support for Python version 3.7 is dropped
* Elements that were deprecated in v0.4 are now removed, as announced

Changed

* Support for `macOS-10.15` is replaced by support for `macOS-12`
* Support for `scikit-learn 0.23` is replaced by support for `scikit-learn 1.0`
* Specify supported `TensorFlow` version (2.9.0)

Added

* Support for Python versions 3.9 and 3.10

Fixed

* Fixed a bug about result of path in `lcss_path_from_metric` function
* Fixed incompatibilities between `NumPy`, `TensorFlow` and `scikit-learn` versions
* Fixed a bug preventing tslearn installation by removing the `NumPy` version constraint (<=1.19) in the file
`pyproject.toml`

Removed

* Cython is now replaced by Numba
* Support for Python versions 3.5 and 3.6 is dropped

Changed

* In docs, change references to `master` branch to `main` branch.

Changed

* Code refactoring to have all subpackages in subfolders
* Improved warnings in `datasets` loading
* `shapelets` module is now compatible with `tensorflow` 2.4

Added

* Added canonical time warping (`ctw` and `ctw_path`)
* `soft_dtw_alignment` provides soft alignment path for soft-dtw
* `lcss` is a similarity measure based on the longest common subsequence
* `lcss_path_from_metric` allows one to pick a dedicated ground metric on top
of which the LCSS algorithm can be run

Fixed

* numpy array hyper-parameters can now be serialized using `to_*()`
methods
* avoid `DivisionByZero` in `MinMaxScaler`
* Fixed incompatibilities with `scikit-learn` 0.24

Removed

* Barycenter methods implemented as estimators are no longer provided: use
dedicated functions from the `tslearn.barycenters` module instead

Changed

* k-means initialization function within `clustering/kmeans.py` updated