国际清算银行-gingado：一个专注于经济和金融的机器学习图书馆（英）-2023.9-30页_252kb

Summary of BIS Working Paper on gingado

Douglas K G Araujo introduces gingado, an open-source Python library designed for machine learning applications in economics and finance. The library is built on three core principles: flexibility (allowing customization), compatibility (seamless integration with libraries like scikit-learn), and responsibility (promoting ethical considerations in model documentation).

Key features of gingado include:

• Data augmentation: Users can easily integrate official statistical data from sources via the SDMX protocol, ensuring data consistency and reproducibility. This helps improve model performance without manual data handling.
• Automatic benchmark models: The library provides ready-to-use random forest models that achieve reasonable performance out-of-the-box, with options to compare or customize them. This simplifies the ML workflow for researchers and practitioners.
• Dataset generation: Includes real and simulated datasets, such as optimized versions of benchmark data, to support testing and causal inference studies.
• Model documentation: Automated tools facilitate comprehensive documentation, incorporating ethical reviews, variable explanations, and other key details, ensuring models are transparent and responsibly deployed.

Gingado is actively developed, aiming to cover areas like clustering, causal ML, and large language model integration. It serves as a tool to streamline ML processes in economics, emphasizing good modeling practices.

This library targets academics and practitioners seeking to leverage ML for economic analysis, with a focus on ethical and reproducible research.