Tsururu: A Python-based Time Series Forecasting Strategies Library

2025-09-19

AI Summary

Current time-series research emphasizes model innovation while neglecting the critical yet underexplored dimension of training strategy design. To address this gap, we propose a modular time-series training strategy framework that formalizes training as a composition of reusable, plug-and-able components—supporting global modeling, multivariate joint forecasting, and multi-step prediction. Based on this framework, we develop TS-Train, an open-source Python library enabling seamless integration and flexible extension with state-of-the-art models (e.g., N-BEATS, TSMixer). Our approach is the first to systematically decouple training strategies into interoperable modules, significantly enhancing strategy reusability and cross-model adaptability. Extensive experiments demonstrate consistent accuracy improvements across multiple benchmark datasets. Furthermore, the framework has been successfully deployed in industrial applications, lowering the technical barrier for complex time-series modeling. The implementation is publicly available on GitHub.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Multi-modal Vision
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

While current time series research focuses on developing new models, crucial questions of selecting an optimal approach for training such models are underexplored. Tsururu, a Python library introduced in this paper, bridges SoTA research and industry by enabling flexible combinations of global and multivariate approaches and multi-step-ahead forecasting strategies. It also enables seamless integration with various forecasting models. Available at https://github.com/sb-ai-lab/tsururu .

Problem

• Selecting optimal training strategies for time series models
• Bridging gap between SoTA research and industry applications
• Enabling flexible forecasting approaches and model integration

Innovation

• Python library for flexible time series forecasting
• Combines global and multivariate forecasting approaches
• Seamless integration with various forecasting models