Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
In the field of choice modeling, the availability of ever-larger datasets has the potential to significantly expand our understanding of human behavior, but this prospect is limited by the poor scalability of discrete choice models (DCMs): as sample sizes increase, the computational cost of maximum likelihood estimation quickly becomes intractable for anything but trivial model structures. To tackle this issue, this study builds upon the work of Lederrey et al. (2021) and the adaptive batch size algorithm they propose for the estimation of DCMs. Specifically, we investigate the use of a dataset reduction technique to generate weighted batches that better represent the whole dataset and, as a result, lead the optimization algorithm to faster convergence. We use a real-world dataset and models of different sizes to compare the performance of our approach with existing methods used in practice.