Mercury: Reusable and Efficient ML Workflows in Finance

Published: 30 Oct 2024, Last Modified: 11 Nov 2024ACM ICAIF P2P Workshop 2024 OralEveryoneRevisionsBibTeXCC BY 4.0
Keywords: Finance, machine learning, opensource
TL;DR: Mercury is a Python library enhancing collaboration and rapid prototyping for predictive models in financial applications. With a modular, open-source architecture, it streamlines development through schema management, explainability, and monitoring
Abstract: Mercury is a Python library developed at BBVA specifically designed to enhance collaboration among data scientists by enabling the efficient sharing of code, thereby addressing inefficiencies in the development and deployment of predictive models for financial applications. Featuring a modular architecture, Mercury ensures adaptability and ease of use, while significantly enhancing the development process of analytical models. The library's evolution from innersource to open-source has broadened its accessibility and fostered a global community of users and contributors. This paper discusses the main modules in Mercury, including advanced schema management, extensive model and data testing, event prediction analysis, model explainability, and mechanisms for monitoring data and model drift. Together, these components streamline the end-to-end ML model development process, providing a robust solution to the challenges of rapid deployment and scalability in a competitive environment.
Submission Number: 2
Loading