Time Series Chains

This is the supporting webpage of the ICDM paper "Matrix Profile VII: Time Series Chains: A New Primitive for Time Series Data Mining" by Yan Zhu, Makoto Imamura, Daniel Nikovski and Eamonn Keogh.

Paper

The paper is here.

Code

The Matlab code can be found here.

Case Studies

The tilt table data can be found here. The data is adapted from [1].

The penguin data can be found here. A short snippet of the data is here.

The human gait data can be found here. Subsequence length is 50. To see the result, run:

>>TSC1_demo( VarName3(1:end-30) ,50);

The web query data can be found here. Note that to make the plot more readable in this undersampled dataset, we upsampled by four, and smoothed the data, like this:

>> x= reshape( [VarName1 ,VarName1,VarName1,VarName1]', 2*1100,1 );

>> TSC1_demo(smooth(x,4),4*19);

However, we get the same basic result in the raw space too.

Robustness Test

The complete code to generate Figure 15 can be found here. The random seed is here. To see the result, run:

>>[ChainLength Svec]=TestChain_Noise(50,20,ChainLength_50_20_sintorandomwalk_seed);

>>figure;plot(1:100,mean(ChainLength,1));

>>figure;plot(1:100,mean(Svec,1));

Chain Calibration Datasets

Here are the datasets and instructions that one can use to further test the robustness of the chain discovery algorithm.

References

[1] Heldt T, Oefinger MB, Hoshiyama M, Mark RG. Circulatory response to passive and active changes in posture. Comput Cardiol, 30:263–266, Sept. 2003.