Change Log¶

Version 0.13.0¶

Add support for Python 3.10 and 3.11, and drop support for Python 3.7.
Update the minimal versions required of the dependencies: * NumPy (>= 1.22.4) * SciPy (>= 1.8.1) * Scikit-Learn (>=1.2.0) * Joblib (>=1.1.1) * Numba (>=0.55.2)
Add an example illustrating time series clustering using pyts.transformation.BOSS transformation with different metrics (by Lucas Plagwitz).
Add automatic components-grouping in the Singular Spectrum Analysis for trend-seasonal decomposition with suitable example (by Lucas Plagwitz).
Add two new parameters in pyts.decomposition.SingularSpectrumAnalysis: chunksize allows for computing the decomposition of all the input time series using chunks (it should be a bit slower but use less memory), and n_jobs allows for running the decomposition of each chunk in parallel.
Set the number of initiations of K-means to compute the initial shapelets in pyts.classification.LearningShapelets: to 10 (to prevent a change of the default value in scikit-learn).
Replace base_estimator_ attribute with estimator_ in pyts.classification.TimeSeriesForest and pyts.classification.TSBF (to match the changes made in scikit-learn).

Add support for Python 3.9 and drop support for Python 3.6.
Add the Time Series Forest algorithm implemented as pyts.classification.TimeSeriesForest.
Add the Time Series Bag-of-Features algorithm implemented as pyts.classification.TSBF.
Replace scikit-learn mixin classes with pyts mixin classes to have standardized docstrings.
Update the examples in the Imaging time series section of the gallery of examples.
Remove some constraints when discretizing time series (number of bins, time series with low variance) that impact the following classes:
Remove specific functions for the different variants of Dynamic Time Warping (all dtw_* functions), only the main pyts.metrics.dtw() is kept.

Add support for Python 3.8 and drop support for Python 3.5.
Rework the BagOfWords algorithm to match the description of the algorithm in the original paper. The former version of BagOfWords is available as WordExtractor in the pyts.bag_of_words module.
Update the SAXVSM classifier with the new version of BagOfWords.
Add the BagOfPatterns algorithm in the pyts.transformation module.
Add the ROCKET algorithm in the pyts.transformation module.
Add the LearningShapelets algorithm in the pyts.classification module.
Deprecated specific functions for Dynamic Time Warping (all dtw_* functions), only the main pyts.metrics.dtw() is kept.

Adapt DTW functions to compare time series with different lengths (by Hicham Janati)
Add a precomputed_cost parameter in DTW variants that are compatible with a precomputed cost matrix, that is classical DTW and DTW with global constraint regions like Sakoe-Chiba band and Itakura parallelogram (by Hicham Janati)
Add a new algorithm called ShapeletTransform in the pyts.transformation module.
Add a new dependency, the joblib Python package, since it has been vendored from scikit-learn and it is used in ShapeletTransform.
[DOC] Revamp documentation in most sections:
- User guide is much more detailed
- A Scikit-learn compatibility page has been added to highlight the compatibility of pyts estimators with scikit-learn tools like model selection and pipelines.
- A Reproducibility page has been added to highlight the work done in the pyts-repro repository, where we compare the performance of our implementations to the literature.
- A Contributing guide has been added.

Add datasets module with dataset loading utilities
Add multivariate module with utilities for multivariate time series
Revamp the tests using pytest.mark.parametrize
Add an Examples section in most of the public functions and classes
Require version 1.3.0 of scipy: this is required to load ARFF files with relational attributes using scipy.io.arff.loadarff