Tslearn

and this project adheres to
[Semantic Versioning](https://semver.org/spec/v2.0.0.html).

Changelogs for this project are recorded in this file since v0.2.0.

* Better initialization schemes for `TimeSeriesKMeans` that lead to more
consistent clustering runs (helps avoid empty cluster situations)
* `TimeSeriesScalerMeanVariance` and `TimeSeriesScalerMinMax` are now
completely sklearn-compliant
* The `shapelets` module now requires tensorflow>=2 as dependency (was keras

Changed

* The structure of the class `Backend` has been simplified.

Added

* Option `compute_with_backend` in functions `soft_dtw`, `soft_dtw_alignment`,
`cdist_soft_dtw` and `cdist_soft_dtw_normalized`.
`PyTorch` automatic differentiation can now be used when these functions are using the `PyTorch` backend.

Fixed

* Fixed error in `LearningShapelets` when input parameter `n_shapelets_per_size` equals `None`.
* Fixed bug related to `SoftDTWLossPytorch` with option `normalize=True` when used on inputs of different lengths.
* Fixed error in function `from_hdf5` for array inputs.
* Fixed `readthedocs` test failing by replacing `build.image` (deprecated) with `build.os`.

Fixed

* Fixed an incorrect calculation of the normalization term for `cdist_soft_dtw_normalized` when `dataset2` is provided.
* Fixed UCR/UEA datasets download link

Fixed

* Fixed an import error when `torch` is not installed. This error appeared in tslearn v0.6.
`PyTorch` is now an optional dependency.