ChronoEpilogi: Scalable time-series variables Selection with Multiple Solutions

Codebase for the experiments of the paper ChronoEpilogi: Scalable Time Series Selection with Multiple Solutions, NeurIPS, 2024, https://doi.org/10.52202/079017-4267

This repository is not maintained, and corresponds to the state of ChronoEpilogi at the time of publication.

Instead, we provide an upgraded version of ChronoEpilogi as a pip installable package: https://github.com/ev07/minimal_chronoepilogi

The documentation for this package is available here: https://ev07.github.io/minimal_chronoepilogi/

Abstract

We consider the problem of selecting all the minimal-size subsets of multivariate time-series (TS) variables whose past leads to an optimal predictive model for the future (forecasting) of a given target variable (multiple feature selection problem for times-series). Identifying these subsets leads to gaining insights, domain intuition, and a better understanding of the data-generating mechanism; it is often the first step in causal modeling. While identifying a single solution to the feature selection problem suffices for forecasting purposes, identifying all such minimal-size, op- timally predictive subsets is necessary for knowledge discovery and important to avoid misleading a practitioner.

We develop the theory of multiple feature selection for time-series data, propose the ChronoEpilogi algorithm, and prove its soundness and completeness under two mild, broad, non-parametric distributional assumptions, namely Compositionality of the distribution and Interchangeability of time-series variable in solutions. Exper- iments on synthetic and real datasets demonstrate the scalability of ChronoEpilogi to thousands of TS variables and its efficacy in identifying multiple solutions. In the real datasets, ChronoEpilogi is shown to reduce the number of TS variables by 96% (on average) without a significant drop in forecasting performance. Furthermore, ChronoEpilogi is on par with group Lasso performance, with the added benefit of providing multiple solutions.

Requirements

pip install -r requirements.txt

Exemple Usage

Synthetic MTS are generated by runing the notebook in ./data/equivalence_datasets.

Tuning is launched with

./tuning/tuning.sh

and Testing with

./testing/testing.sh.

Code structure

• Implementation of the proposed algorithm is in the main directory, under ChronoEpilogi.py. Subroutines are implemented in associations.py and models.py.
• Baselines and Forecasters are wrapped respectively in \baselines\feature_selection.py and \baselines\estimators.py. Wrappers allows for standard call structure and easier pipeline declaration across libraries.
• Datasets generation assets are provided in the \data\ folder. Dataset opening routine is implemented in data_opener.py
• Tuning of the algorithms is implemented in folder \tuning\.
  • first_wave_main.py implements paired tuning of feature selection algorithm and forecaster.
  • second_wave_main.py implements tuning of a forecaster for an already tuned feature selection algorithm.
  • routines.py implement shared elements between tuning routines.
  • final_statistics.py implement tuning record opening routines.
• Final evaluation on a test set is implemented in folder \testing\:
  • results.py and results_other.py implement respectively ChronoEpilogi multisolution evaluation and other fs algorithms evaluation.
  • final_statistics.py implement tuning record opening routines.

Contributing

Our dataset is open source and free of use, under the licence provided in LICENSE.