TFB: An Open-Source Machine Learning Library Designed for Time Series Researchers

TFB is an open-source machine learning library designed to facilitate the empirical evaluation and comparison of time series forecasting methods. It addresses limitations in existing benchmarks by offering a comprehensive dataset collection, diverse evaluation strategies and metrics, and a flexible and scalable pipeline. Experimentation using TFB has provided practical insights into the performance of different time series forecasting methods across various datasets and characteristics.