SI2-SSE: Improving Scikit-Learn Usability and Automation

Columbia University, New York NY

Investigators

Abstract

Machine learning is a central component in many data-driven research areas, but its adoption is limited by the often complex choice of data processing, model, and parameter settings. The goal of this project is to create software tools that enable automatic machine learning, that is, solving predictive analytics tasks without requiring the user to explicitly specify the algorithm or model parameters used for prediction. The software developed in this project will enable a wider use of machine learning, by providing tools to apply machine learning without requiring knowledge of the details of the algorithms involved. The project, supported by the Office of Advanced Cyberinfrastructure, and the Division of Computing and Communication Foundations, extends the existing scikit-learn project, a machine learning library for Python, which is widely used in academic research across disciplines. The project will add features to this library to lower the amount of expert knowledge required to apply models to a new problem, and to facilitate the interaction with automated machine learning systems. The project will also create a separate software package that includes models for automatic supervised learning, with a very simple interface, requiring minimal user interaction. In contrast to existing research projects, this project focuses on creating easy-to-use tools that can be used by researchers without extensive training in machine learning or computer science.