The results reported in [4] were a version of the code that has since changed. We use a more recent version of Weka and we have altered the shapelet implementation to include the speedups described in the original work [1], [2].

A key implementation issue is in what order to evaluate the series. This will influence the usefulness of the early abandon and can also affect the quality of the shapelets found (depending on how ties in quality measure are resolved). We have adopted a round robin ordering, where the series are ordered for evaluation so that class values follow a structured pattern (re.g. 1,2,3,1,2,3).

The full results are in this spreadsheet. Whilst there are differences in accuracy (some of which are surprisingly large), the overall message of the paper is the same. Transforming produces significantly better results on these data sets. There are critical difference diagrams on the spreadsheet that demonstrate this point.

References

[1] Ye, L and Keogh, E., Time series shapelets: a new primitive for data mining Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, 2009.
[2] Mueen, A, Keogh, E and Young, N, Logical-shapelets: An expressive primitive for time series classification Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining, 2011.
[3] Rakthanmanon, T , Campana, b, Mueen, a, Batista, g, Westover, m, Zhu, Q , Zakaria, Keogh, E. Searching and mining trillions of time series subsequences under dynamic time warping. Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining
[4] Hills, J, Lines, J, Baranauskas, E, Mapp, J, and Bagnall, A., Time Series Classification with Shapelets, Journal of Data Mining and Knowledge Discovery. online first preprint.pdf, 2013.