tslearn is a Python package that provides machine learning tools for the analysis of time series. This package builds on scikit-learn, numpy and scipy libraries.

If you would like to contribute to tslearn, please have a look at our contribution guidelines.

Cython
numpy
numba
scipy
scikit-learn
joblib
numba

If you plan to use the shapelets module, keras and tensorflow should also be installed. h5py is required for reading or writing models using the hdf5 file format.

The easiest way to install tslearn is probably via conda:

conda install -c conda-forge tslearn

When using PyPI, C++ build tools should be available to perform installation.

Using pip should also work fine:

pip install tslearn

In this case, you should have numpy, cython and C++ build tools available at build time.

If you want to get tslearn's latest version, you can refer to the repository hosted at github:

pip install git+https://github.com/rtavenar/tslearn.git

It seems on some platforms Cython dependency does not install properly. If you experiment such an issue, try installing it with the following command:

pip install cython

before you start installing tslearn. If it still does not work, we suggest you switch to conda installation.

The documentation, including a gallery of examples, is hosted at readthedocs.

• A generators module provides Random Walks generators
• A datasets module provides access to the famous UCR/UEA datasets through the UCR_UEA_datasets class
• A preprocessing module provides standard time series scalers
• A metrics module provides:
  • Dynamic Time Warping (DTW) (with Sakoe-Chiba band and Itakura parallelogram variants)
  • LB_Keogh
  • Global Alignment Kernel
  • Soft-DTW from Cuturi and Blondel
• A neighbors module includes nearest neighbor algorithms to be used with time series
• An svm module includes Support Vector Machine algorithms with:
  • Standard kernels offered in sklearn (with adequate array reshaping done for you)
  • Global Alignment Kernel
• A clustering module includes the following time series clustering algorithms:
  • Standard Euclidean k-means (with adequate array reshaping done for you)
    • Based on tslearn.barycenters
  • DBA k-means from Petitjean et al.
    • Based on tslearn.barycenters that offers DBA facility that could be used for other applications than just k-means
  • Global Alignment kernel k-means
  • KShape clustering from Paparizzos and Gravano
  • Soft-DTW k-means from Cuturi and Blondel
    • Based on tslearn.barycenters that offers Soft-DTW barycenter computation
  • It also provides a way to compute the silhouette coefficient for given clustering and metric
• A shapelets module includes an efficient implementation of the Learning Time-Series method from Grabocka et al.
  • Warning: to use the shapelets module, two extra dependencies are required: keras and tensorflow
• A piecewise module includes standard time series transformations, as well as the corresponding distances:
  • Piecewise Aggregate Approximation (PAA)
  • Symbolic Aggregate approXimation (SAX)
  • 1d-Symbolic Aggregate approXimation (1d-SAX)

Have a look there for a list of suggested features. If you want other ML methods for time series to be added to this TODO list, do not hesitate to open an issue! See our contribution guidelines for more information about how to proceed.

Authors would like to thank Mathieu Blondel for providing code for Kernel k-means and Soft-DTW (both distributed under BSD license) that are used in the clustering and metrics modules of this library.

If you use tslearn in a scientific publication, we would appreciate citations:

@misc{tslearn,
      title={tslearn: A machine learning toolkit dedicated to time-series
             data},
      author={Romain Tavenard and Johann Faouzi and Gilles Vandewiele and
              Felix Divo and Guillaume Androz and Chester Holtz and Marie
              Payne and Roman Yurchak and Marc Ru{\ss}wurm and Kushal
              Kolar and Eli Woods},
      year={2017},
      note={\url{https://github.com/rtavenar/tslearn}}
}