Multiple-Resolution Tokenization for Time Series Forecasting with an Application to Pricing

TMLR Paper2954 Authors

Abstract: We propose a transformer architecture for time series forecasting with a focus on time series tokenisation and apply it to a real-world prediction problem from the pricing domain. Our architecture aims to learn effective representations at many scales across all available data simultaneously. The model contains a number of novel modules: a differentiated form of time series patching which employs multiple resolutions, a multiple-resolution module for time-varying known variables, a mixer-based module for capturing cross-series information, and a novel output head with favourable scaling to account for the increased number of tokens. We present an application of this model to a real world prediction problem faced by the markdown team at a very large retailer. On the experiments conducted our model outperforms in-house models and the selected existing deep learning architectures.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

1. Performed three ablation experiments (choice of resolution set, inclusion of modules, scaling) on an analogous public retail dataset (Favorita Grocery Sales Forecasting).
2. Clarified the difference between our approach and the multi scale approaches mentioned by the reviewers (multi-fidelity, neural operators, and mixed-resolution).
3. Formed a clearer justification for the use of transformers for this specific forecasting problem.
4. Added a motivation for the novel output head design.
5. Changed the paper structure and edited some of the writing to make the paper cleaner and more focused on core contributions.

Assigned Action Editor: ~Nicolas_THOME2

Submission Number: 2954