TSInterpret: A unified framework for time series interpretability

Abstract

With the increasing application of deep learning algorithms to time series classification, especially in high-stake scenarios, the relevance of interpreting those algorithms becomes key. Although research in time series interpretability has grown, accessibility for practitioners is still an obstacle. Interpretability approaches and their visualizations are diverse in use without a unified api or framework. To close this gap, we introduce TSInterpret 1 , an easily extensible open-source Python library for interpreting predictions of time series classifiers that combines existing interpretation approaches into one unified framework. The library features (i) state-of-the-art interpretability algorithms, exposes a (ii) unified API enabling users to work with explanations in a consistent way, and provides (iii) suitable visualizations for each explanation.

Explainability of CNN-based Alzheimer’s Disease Detection from Online Handwriting

“…To implement the CoMTE method, we utilized the TSInterpret library [24], which offers a comprehensive and user-friendly framework for generating explanations for time series data. Figure 2 shows several examples of CoMTE output for loops of different subjects.…”

“…We used the TSInterpret library [24] to implement the 2-step TSR method. Given an input instance, TSR returns normalized time slices and feature importance scores in range [0, 1].…”

“…Furthermore, stateof-the-art research on the explainability of multivariate time series have converged on the efficacy of two specialized local methods: the Counterfactual explanation-based method (Comte) and Two-Step Temporal Saliency Rescaling (TSR) [24] [22] [21] [23]. These methods, purpose-built for multivariate time series data, have showcased exceptional results and have their implementations readily accessible in the TSInterpret library [24]. In our work, we focus on three distinct methods: DeepSHAP [19], offering both local and global explanations, CoMTE [21], a sample-based local explainability method providing counterfactual explanations, and 2-Step TSR [23], a feature-based local explanation method.…”