Merlion: End-to-End Machine Learning for Time Series
Aadyot Bhatnagar, Paul Kassianik, Chenghao Liu, Tian Lan, Wenzhuo Yang, Rowan Cassius, Doyen Sahoo, Devansh Arpit, Sri Subramanian, Gerald Woo, Amrita Saha, Arun Kumar Jagota, Gokulakrishnan Gopalakrishnan, Manpreet Singh, K C Krithika, Sukumar Maddineni, Daeki Cho, Bo Zong, Yingbo Zhou, Caiming Xiong, Silvio Savarese, Steven Hoi, Huan Wang; 24(226):1−6, 2023.
Abstract
Merlion is a time series machine learning library that is open-source. It provides a unified interface for commonly used models and datasets for forecasting and anomaly detection in both univariate and multivariate time series. Additionally, it includes standard pre/post-processing layers. The library offers various modules to enhance user-friendliness, such as a no-code visual dashboard, anomaly score calibration for improved interpretability, AutoML for hyperparameter tuning and model selection, and model ensembling. Merlion also provides an evaluation framework that simulates the live deployment of a model in production, as well as a distributed computing backend for running time series models at an industrial scale. The goal of this library is to provide engineers and researchers with a comprehensive solution for rapidly developing models tailored to their specific time series requirements and benchmarking them across multiple datasets.
[abs]