Went to a seminar today on the reproducibility crisis in ML science (Reproducibility workshop). It had some really great talks, but in particular the final couple by Jessica Hullman and Brandon Stewart (starting at about 5:00:00) who talked about issues around reproducible workflows and causality in ML modelling that may be of interest to the group. The police DAG example in chapter 3 even came up as an example in Brandon’s talk!
Seminar: The Reproducibility Crisis in ML-based science - YouTube