Installation
There are different alternatives to install tslearn:
- PyPi:
python -m pip install tslearn
- Conda:
conda install -c conda-forge tslearn
- Git:
python -m pip install https://github.com/tslearn-team/tslearn/archive/main.zip
In order for the installation to be successful, the required dependencies must be installed. For a more detailed guide on how to install tslearn, please see the Documentation.
Getting started
1. Getting the data in the right format
tslearn expects a time series dataset to be formatted as a 3D numpy array. The three dimensions correspond to the number of time series, the number of measurements per time series and the number of dimensions respectively (n_ts, max_sz, d). In order to get the data in the right format, different solutions exist:
It should further be noted that tslearn supports variable-length timeseries.
>>> from tslearn.utils import to_time_series_dataset
>>> my_first_time_series = [1, 3, 4, 2]
>>> my_second_time_series = [1, 2, 4, 2]
>>> my_third_time_series = [1, 2, 4, 2, 2]
>>> X = to_time_series_dataset([my_first_time_series,
my_second_time_series,
my_third_time_series])
>>> y = [0, 1, 1]
2. Data preprocessing and transformations
Optionally, tslearn has several utilities to preprocess the data. In order to facilitate the convergence of different algorithms, you can scale time series. Alternatively, in order to speed up training times, one can resample the data or apply a piece-wise transformation.
>>> from tslearn.preprocessing import TimeSeriesScalerMinMax
>>> X_scaled = TimeSeriesScalerMinMax().fit_transform(X)
>>> print(X_scaled)
[[[0.] [0.667] [1.] [0.333] [nan]]
[[0.] [0.333] [1.] [0.333] [nan]]
[[0.] [0.333] [1.] [0.333] [0.333]]]
3. Training a model
After getting the data in the right format, a model can be trained. Depending on the use case, tslearn supports different tasks: classification, clustering and regression. For an extensive overview of possibilities, check out our gallery of examples.
>>> from tslearn.neighbors import KNeighborsTimeSeriesClassifier
>>> knn = KNeighborsTimeSeriesClassifier(n_neighbors=1)
>>> knn.fit(X_scaled, y)
>>> print(knn.predict(X_scaled))
[0 1 1]
As can be seen, the models in tslearn follow the same API as those of the well-known scikit-learn. Moreover, they are fully compatible with it, allowing to use different scikit-learn utilities such as hyper-parameter tuning and pipelines.
4. More analyses
tslearn further allows to perform all different types of analysis. Examples include calculating barycenters of a group of time series or calculate the distances between time series using a variety of distance metrics.
Available features
Documentation
The documentation is hosted at readthedocs. It includes an API, gallery of examples and a user guide.
Contributing
If you would like to contribute to tslearn, please have a look at our contribution guidelines. A list of interesting TODO's can be found here. If you want other ML methods for time series to be added to this TODO list, do not hesitate to open an issue!
Referencing tslearn
If you use tslearn in a scientific publication, we would appreciate citations:
@article{JMLR:v21:20-091,
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},
title = {Tslearn, A Machine Learning Toolkit for Time Series Data},
journal = {Journal of Machine Learning Research},
year = {2020},
volume = {21},
number = {118},
pages = {1-6},
url = {https://www.jmlr.org/papers/v21/20-091.html}
}
Acknowledgments
Authors would like to thank Mathieu Blondel for providing code for Kernel k-means and Soft-DTW, and to Mehran Maghoumi for his torch-compatible implementation of SoftDTW.
