CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

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

[Towards v0.8.0]

Changed

• Rework cdist_dtw and underlying Numba jited DTW computation to speed things up for numpy Backends.
• Masks as returned by compute_mask, sakoe_chiba_mask and itakura_mask now have boolean values.
• Backend attributes are now dynamic.
• Ensure dtype is preserved in to_time_series_dataset
• Fixed centroids computations in K-shape for multivariate timeseries (#288)
• Fixed json and pickle serialization for LearningShapelet (#387)(#403)

Added

• DBSCAN estimator for timeseries (#598)

[v0.7.0]

Changed

• Explicit exception when using Global Alignment Kernel with sigma close to zero (#440)
• Fixed shifting in K-shape shape extraction process (#385)
• Support for scikit-learn up to 1.7 (#549)
• Fixed LearningShapelets with variable length timeseries (#352)
• The shapelets module now depends on Keras3+ and the underlying backend can be selected through the KERAS_BACKEND environment variable. Defaults to the first found installed backend among torch, tensorflow and jax (in that order).

Removed

• Support for Python versions 3.8 and 3.9 is dropped

Added

• per_timeseries and per_feature options for min-max and mean-variance scalers (#536)
• TimeSeriesImputerclass: missing value imputer for time series (#564)
• Frechet metrics and KNeighbors integration (#402)

[v0.6.4]

Changed

• Support for scikit-learn up to 1.6

[v0.6.3]

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.

[v0.6.2]

Fixed

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

[v0.6.1]

Fixed

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

[v0.6.0]

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

[v0.5.3]

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

[v0.5.2]

Changed

• In docs, change references to master branch to main branch.

[v0.5.0]

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

[v0.4.0]

Changed

• k-means initialization function within clustering/kmeans.py updated to be compatible with scikit-learn 0.24
• 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 tensorflow==1.* up to version 0.3)
• GlobalAlignmentKernelKMeans is deprecated in favor of KernelKMeans that accepts various kernels (and "gak" is the default)
• ShapeletModel is now called LearningShapelets to be more explicit about which shapelet-based classifier is implemented. ShapeletModel is still available as an alias, but is now considered part of the private API

Added

• Python 3.8 support
• dtw_path_from_metric allows one to pick a dedicated ground metric on top of which the DTW algorithm can be run
• Nearest Neighbors on SAX representation (with custom distance)
• Calculate pairwise distance matrix between SAX representations
• PiecewiseAggregateApproximation can now handle variable lengths
• ShapeletModel is now serializable to JSON and pickle formats
• Multivariate datasets from the UCR/UEA archive are now available through UCR_UEA_datasets().load_dataset(...)
• ShapeletModel now accepts variable-length time series dataset; a max_size parameter has been introduced to save room at fit time for possibly longer series to be fed to the model afterwards
• ShapeletModel now accepts a scale parameter that drives time series pre-processing for better convergence
• ShapeletModel now has a public history_ attribute that stores loss and accuracy along fit epochs
• SAX and variants now accept a scale parameter that drives time series pre-processing to fit the N(0,1) underlying hypothesis for SAX
• TimeSeriesKMeans now has a transform method that returns distances to centroids
• A new matrix_profile module is added that allows MatrixProfile to be computed using the stumpy library or using a naive "numpy" implementation.
• A new early_classification module is added that offers early classification estimators
• A new neural_network module is added that offers Multi Layer Perceptron estimators for classification and regression

Fixed

• Estimators that can operate on variable length time series now allow for test time datasets to have a different length from the one that was passed at fit time
• Bugfix in kneighbors() methods.

Removed

• Support for Python 2 is dropped

[v0.3.1]

Fixed

• Fixed a bug in TimeSeriesSVC and TimeSeriesSVR that caused user-input gamma to be ignored (always treated as if it were "auto") for gak kernel

[v0.3.0]

Changed

• dtw_barycenter_averaging is made faster by using vectorized computations
• dtw_barycenter_averaging can be restarted several times to reach better local optima using a parameter n_init set to 1 by default
• Functions load_timeseries_txt and save_timeseries_txt from the utils module have changed their names to load_time_series_txt and save_time_series_txt. Old names can still be used but considered deprecated and removed from the public API documentation for the sake of harmonization
• Default value for the maximum number of iterations to train ShapeletModel and SerializableShapeletModel is now set to 10,000 (used to be 100)
• TimeSeriesScalerMeanVariance and TimeSeriesScalerMinMax now ignore any NaNs when calling their respective transform methods in order to better mirror scikit-learn's handling of missing data in preprocessing.
• KNeighborsTimeSeries now accepts variable-length time series as inputs when used with metrics that can deal with it (eg. DTW)
• When constrained DTW is used, if the name of the constraint is not given but its parameter is set, that is now considered sufficient to identify the constraint.

Added

• KNeighborsTimeSeriesRegressor is a new regressor based on k-nearest-neighbors that accepts the same metrics as KNeighborsTimeSeriesClassifier
• A set_weights method is added to the ShapeletModel and
  SerializableShapeletModel estimators
• subsequence_path and subsequence_cost_matrix are now part of the public API and properly documented as such with an example use case in which more than one path could be of interest (cf. plot_sdtw.py)
• verbose levels can be set for all functions / classes that use joblib for parallel computations and joblib levels are used;
• conversion functions are provided in the utils module to interact with other Python time series packages (pyts, sktime, cesium, seglearn, tsfresh, stumpy, pyflux)
• dtw_barycenter_averaging_subgradient is now available to compute DTW barycenter based on subgradient descent
• dtw_limited_warping_length is provided as a way to compute DTW under upper bound constraint on warping path length
• BaseModelPackage is a base class for serializing models to hdf5, json and pickle. h5py is added to requirements for hdf5 support.
• BaseModelPackage is used to add serialization functionality to the following models: GlobalAlignmentKernelKMeans, TimeSeriesKMeans, KShape, KNeighborsTimeSeries, KNeighborsTimeSeriesClassifier, PiecewiseAggregateApproximation, SymbolicAggregateApproximation, and OneD_SymbolicAggregateApproximation

[v0.2.4]

Fixed

• The tests subdirectory is now made a python package and hence included in wheels

[v0.2.2]

Fixed

• The way version number is retrieved in setup.py was not working properly on Python 3.4 (and made the install script fail), switched back to the previous version

[v0.2.1]

Added

• A RuntimeWarning is raised when an 'itakura' constraint is set that is unfeasible given the provided shapes.

Fixed

• 'itakura' and 'sakoe_chiba' were swapped in metrics.compute_mask

[v0.2.0]

Added

• tslearn estimators are now automatically tested to match sklearn requirements "as much as possible" (cf. tslearn needs in terms of data format, etc.)
• cdist_dtw and cdist_gak now have a n_jobs parameter to parallelize distance computations using joblib.Parallel
• n_jobs is also available as a prameter in silhouette_score, TimeSeriesKMeans, KNeighborsTimeSeries, KNeighborsTimeSeriesClassifier, TimeSeriesSVC, TimeSeriesSVR and GlobalAlignmentKernelKMeans

Changed

• Faster DTW computations using numba
• tslearn estimators can be used in conjunction with sklearn pipelines and cross-validation tools, even (for those concerned) with variable-length data
• doctests have been reduced to those necessary for documentation purposes, the other tests being moved to tests/*.py
• The list of authors for the tslearn bibliographic reference has been updated to include Johann Faouzi and Gilles Van de Wiele
• In TimeSeriesScalerMinMax, min and max parameters are now deprecated in favor of value_range. Will be removed in v0.4
• In TimeSeriesKMeans and silhouette_score, 'gamma_sdtw' is now deprecated as a key for metric_params in favor of gamma. Will be removed in v0.4

Removed

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