xiaoshengli / LL

Code for "Linear Time Motif Discovery in Time Series"

Linear Time Motif Discovery in Time Series (Li-Lin algorithm, LL)

This repository contains the code accompanying the paper, "Linear Time Motif Discovery in Time Series" (Xiaosheng Li and Jessica Lin, SDM 2019). This paper presents an algorithm that can find the exact motif of a given time series in a linear expected time complexity. The algorithm is further modified to find all pairs of subsequences whose distances are below a given threshold value.

The code can also be used to find the closest pair among a set of data points with slight modification. The code is considered suitable for the situation where the data points have high dimension.

To Compile the Code

Assume using a Linux system:

g++ -O3 -o LL LL.cpp -std=c++11

g++ -O3 -o LL2 LL2.cpp -std=c++11

To Run the Code

./LL [timeSeriesFile] [timeSeriesLength] [motifLength] [g]

[timeSeriesFile] is the name of the time series file to run, the user needs to place the file in the same directory as LL, [timeSeriesLength] is the length of the input time series, [motifLength] is the motif length, [g] is the number of grids. [g] supports cascading setting up to 3 values.

./LL2 [timeSeriesFile] [timeSeriesLength] [motifLength] [threshold] [g]

[timeSeriesFile] is the name of the time series file to run, the user needs to place the file in the same directory as LL2, [timeSeriesLength] is the length of the input time series, [motifLength] is the motif length, [threshold] is the distance threshold for the motif to find, [g] is the number of grids.

Examples

./LL data.txt 1000000 10 4

Output:

time series length: 1000000 motif length:10
motif distance(Chebyshev): 2.02865e-05
motif: 
630065, 681413
The running time is: 5.31 seconds

./LL data.txt 1000000 10 2 4 6

Output:

time series length: 1000000 motif length:10
motif distance(Chebyshev): 2.02865e-05
motif: 
630065, 681413
The running time is: 12.68 seconds

./LL2 data.txt 1000000 10 0.0001 4

Output:

time series length: 1000000 motif length: 10 threshold: 0.0001 g: 4
number of motif found: 2
motif and distance (Chebyshev): 
330073, 901243, 6.63788e-05
630065, 681413, 2.02865e-05
The running time is: 2.31 seconds

Citation

@inproceedings{li2019linear,
  title={Linear Time Motif Discovery in Time Series},
  author={Li, Xiaosheng and Lin, Jessica},
  booktitle={Proceedings of the 2019 SIAM International Conference on Data Mining},
  pages={136--144},
  year={2019},
  organization={SIAM}
}