Coming soon: time series clustering and time series regression results.

We host two sets of classification results: the published results are archived for legacy purposes to reproduce results for relevant papers. The reference results are generated with recent versions of the aeon toolkit and represent the state of the art.

Lists of dataset names are stored in aeon. The univariate archive currently contains 128 datasets and is also available from UCR . The multivariate archive currently contains 30 datasets, 26 equal length and 4 unequal length.

Lists of datasets are available as follows

from aeon.datasets.tsc_data_lists import univariate2015, univariate, multivariate, univariate_equal_length, multivariate_equal_length
print(" 85 datasets in 2015 version of UCR = ", univariate85)
print(" 128 datasets in 2019 version of UCR = ", univariate)
print(" 112 equal length/no missing UCR dataets = ", univariate_equal_length)
print(" 30 multivariate tsml datasets = ", multivariate)
print(" 26 equal length multivariate tsml datasets = ", multivariate_equal_length)

Results are stored in files by classifier. Results from the 2017 bake off paper are here. The first bake off was conducted on the 85 univariate data sets that were in the archive at the time. We compared 25 classifier and conducted 100 resamples on each dataset. You can recover the results for the default train split or for the averages over resamples in aeon as follows:

from aeon.benchmarking import load_bakeoff_2017
# Returns a 2D array of (datasets,classifiers) with all the classifiers

default_results = load_bakeoff_2017(default_split=True)
average_results 
X, y, meta_data = load_classification("GunPoint")
print(" Shape of X = ", X.shape)
print(" Meta data = ", meta_data)
You can get a

Results from the 2020 multivariate bakeoff paper are here. This bake off in first bakeoff was conducted on the 85 univariate data sets that were in the archive at the time. You can download these directly with the from here using aeon as follows:

Results from the 2021 HIVE-COTE v2.0 paper are here. This introduced the current state of the art algorithm for time series classification. HC" performs best on both univariate and multivariate problems. One way results are presented is through critical difference diagrams. These show the average rank of classifiers over all problems. Solid bars indicate cliques, within which there is no significant difference in rank. Below is the multivariate example. Figure X from the HC2 paper

You can reproduce this CD diagram with the following code in the aeon toolkit.

Results from the Bake off redux are here. This is under review, and has been for a long time. If accepted, we will provide more information and details here. We revisit advances in the field and repeat the 2017 comparison with new algorithms and new data. as yet revisited introduced the current state of the art algorithm for time series classification. HC" performs best on both univariate and multivariate problems. One way results are presented is through critical difference diagrams. These show the average rank of classifiers over all problems. Solid bars indicate cliques, within which there is no significant difference in rank. Below is the multivariate example.

You can reproduce this CD diagram with the following code in the aeon toolkit.

A comparison of ranks for eight best in category classifiers

Critical difference diagram for eight best of class classifiers. Tests are performed with the sign rank test using the Holm correction. Top clique of four classifiers represent the state of the art in Spring 2020.

Dictionary Based

Interval and Spectral Based

Shapelet Based

Distance Based

Deep Learning Based

Hybrids

Legacy results used in the paper the great time series classification bake off are below. These are on the 85 datasets that were in the UCR archive at the time.
To download train/test results, click here
To download results averaged over 100 resamples, click here
If you want results for all individual folds, click here
Choose a tab to view the repositories results