BasicTS is a comprehensive benchmark library and toolkit for time series forecasting, designed for researchers and practitioners. It provides a unified, fair, and scalable platform for reproducing, comparing, and developing time series forecasting models, supporting a wide array of tasks including spatial-temporal and long-term forecasting.

How It Works

BasicTS employs a config-driven pipeline built on EasyTorch, enabling users to define all aspects of model training and evaluation, from data preprocessing to hyperparameter tuning, via configuration files. This approach ensures reproducibility and facilitates easy experimentation with new models by requiring minimal code implementation for custom architectures. The framework supports distributed training across multiple GPUs and nodes, abstracting away hardware complexities.

Quick Start & Requirements

• Installation: Refer to the Getting Started guide for detailed instructions.
• Prerequisites: PyTorch (versions 1.10.0 to 2.3.1 supported), Python. EasyTorch is used as the backend.
• Resources: Supports CPU, GPU, and distributed GPU training.

Highlighted Details

• Supports a wide range of baseline models for spatial-temporal, long-term, and general time series forecasting.
• Includes implementations for numerous datasets, covering traffic, weather, electricity, and more.
• Offers fair performance review through a unified pipeline for model comparison.
• Features a config-based system for controlling all pipeline details, including curriculum learning.

Maintenance & Community

The project is actively developed and welcomes contributions. An official Discord server is available for community support and discussion.

Licensing & Compatibility

The project is licensed under the MIT license, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

While extensive, the README does not explicitly detail specific limitations or known bugs. The project relies on EasyTorch as a backend, which may introduce its own dependencies or learning curve.