| Citekey | - |
| Source Code | own |
| Learning type | unsupervised |
| Input dimensionality | multivariate |
This is a modified version of the grammarviz algorithm. Modifications include the possibility to classify multivariate time series, and general quality-of-life additions.
Furthermore, parameters for additional configuration of output algorithms were added, and the need for post-processing was removed.
The most important parameters are output_mode and multi_strategy.
output_mode specifies the algorithm which will generate the anomaly scores and
multi_strategy specifies which adaption to the multivariate case should be used.
This only applies for time series with more than one dimension.
The univariate implementation uses output_mode of 2.
| Output mode value | algorithm |
|---|---|
| 0 | rule density |
| 1 | discord discovery (RRA) |
| 2 | modified brute-force HOTSAX |
| Multivariate strategy value | algorithm |
|---|---|
| 0 | merge all dimensions |
| 1 | merge correlated dimensions |
| 2 | merge no dimensions |
Pavel Senin, Jessica Lin, Xing Wang, Tim Oates, Sunil Gandhi, Arnold P. Boedihardjo, Crystal Chen, and Susan Frankenstein. 2018. GrammarViz 3.0: Interactive Discovery of Variable-Length Time Series Patterns. ACM Trans. Knowl. Discov. Data 12, 1, Article 10 (February 2018), 28 pages. DOI: https://doi.org/10.1145/3051126
Senin, P., Lin, J., Wang, X., Oates, T., Gandhi, S., Boedihardjo, A.P., Chen, C., Frankenstein, S., Lerner, M., Time series anomaly discovery with grammar-based compression, The International Conference on Extending Database Technology, EDBT 15.