This website is an ongoing project to develop a comprehensive repository for research into time series classification. The domain is owned by Tony Bagnall and maintained by his research group to help promote reproducable research. Unfortunately, we never have enough time to do it justice. If you are interested in sponsoring this website so we can develop it further, please get in touch. If you use the results or code, please cite the latest bake off paper "Matthew Middlehurst, Patrick Schäfer and Anthony Bagnall, "Bake off redux: a review and experimental evaluation of recent time series classification algorithms" ArXiv (under review)
There are two code repositories associated with this website. The Java based, weka compatible toolkit tsml and the python based, sklearn compatible aeon.
We are in the process of refreshing and expanding the results sections, more information to follow.
Download the tsml classification accuracy results for the 112 UCR univariate TSC problems presented in the univariate bake off and the HC2 paper.Download the tsml classification accuracy results for the 26 UEA multivariate TSC problems presented in the multivariate bake off and the HC2 paper.
Alternatively you can just download the results directly from the website in code. Tools for downloading, presenting and reproducing these results are in the following GitHub repo. This is a work in progress, we would welcome contributions.
If you want to donate data, have any queries or problems with any of the datasets or want to give feedback on the website, please raise an issue on the associated Github repo.
The univariate TSC archive was relaunched in 2018 with 128 datasets.
Weka formatted ARFF files (and .txt files) (about 500 MB).
aeon formatted ts files (about 250 MB).
more info The univariate TSC archive can be referenced with this paper.
The multivariate TSC archive was launched with 30 datasets in 2018.
Weka formatted ARFF files (and .txt files) (about 2 GB).
aeon formatted ts files (about 1.5 GB).
more info. The archive can be referenced with this paper.
If you want to use PyTorch with the tsc.com data, there is a repo to help with formatting here. It links to a python package that downloads and prepares the TSC data sets as PyTorch tensors.The package is independent of anyone associated with this website, and we have not tested it, but its a great idea so we are happy to share, and will try using it soon.