Skip to content

Latest commit

 

History

History
50 lines (36 loc) · 2.72 KB

README.md

File metadata and controls

50 lines (36 loc) · 2.72 KB
Raw

Time Series Multivariate and Univariate Local Explanations (TS-MULE)

Repository for the paper "Local Interpretable Model-Agnostic Explanations For Time Series Forecast Models" submitted to AIMLAI 2021. It is a general extension of LIME [1] for univariate and multivariate time series. We therefore extend LIME with a selection of segmentation algorithms for input data. Further, strategies for the perturbation of the segmented input data is presented.

Ideas and Usage

We focus on six different segmentation ideas:

  • uniform segmentation based on a fixed window size
  • exponential segmentation based on a exponentially growing wndow
  • SAX segmentation based on the SAX transformation to symbols
  • Slopes-sorted segmentation based on the gradient of the matrix profile
  • Slopes-not-sorted segmentation based on the gradient of the matrix profile
  • Bins-min segmentation based on a binning variant of the matrix profile
  • Bins-max segmentation based on a binning variant of the matrix profile

And three different replacement strategies:

  • Zero
  • Inverse
  • Mean

Comparison of segmentation algorithms on example time series

Comparison

Workflow

The correpsonding workflow with LIME adapted to time series.
Workflow
Our proposed segmetnation algorithms enable the first step.

Results

We evaluate the previous proposed segmentation algorithms for TS-MULE with a perturbation analysis based on Schlegel et al. [2]. In our paper, we focused on zero perturbation, but we als have the results for zero, inverse-mean, and global-mean.

Results
In general, we advise to use uniform or one of the bins (e.g., bins-min) for short window lengths.
For larger window lengths, we suggest slopes as it works best, but also the bins (e.g., bins-min) work quite well.

A larger evalaution for the replacement approaches is planned but not yet done.

Acknowledgments

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 826494.

References

[1]: Ribeiro, M. T., Singh, S., & Guestrin, C. (2016, August). "Why should i trust you?" Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining (pp. 1135-1144).
[2]: Schlegel, U., Arnout, H., El-Assady, M., Oelke, D., & Keim, D. A. (2019). Towards A Rigorous Evaluation Of XAI Methods On Time Series. In 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW) (pp. 4321-4325).