I’m happy to share that the first beta version of our Python library, PyDCML, has been released: https://mlsm.man.dtu.dk/pydcml/intro.html

PyDCML is a Python library for fast implementation and scalable inference of Bayesian Discrete Choice Models that makes it easy to leverage flexible state-of-the-art modelling techniques from Machine Learning, while remaining interpretable and preserving the links with economic theories established by Daniel McFadden [Ref1].

PyDCML uses PyTorch on the backend in order to enable stochastic backpropagation, automatic differentiation and GPU-accelerated computation. In doing so, PyDCML aims at enabling flexible and expressive Choice Modeling, unifying the best of modern Machine Learning and Bayesian modeling with Discrete Choice Theory.