Practical iterative model building with stan

I am interested in what concepts (and code) other people use for iterative model building with stan, especially with regard to reproducibility (avoiding rerunning the same model), changing data during the project (new data but also cleaning of existing data), automated posterior predictive checks and automated model-comparison. The paper ‘Bayesian workflow’ by Gelman et al. ([2011.01808] Bayesian Workflow) covers these topics in a conceptual way, but I feel there is room for discussion and tools helping with the practical implementation. Currently, I don’t know of good ‘tools’ helping with these task. Even though it may be impossible to design a readily applicable software tool covering all use cases, it might be possible to collect code examples and implementation approaches.

Are other people interested in having a conversation on this here in the forum? To initiate such a conversation it is probably best to post ‘use cases’ i.e. some information on the modelling project and a short description of the workflow. I will shortly make an example in a separate post.

I am currently at the BayesComp conference in Levi, Finland, and I would be very happy to chat if somebody reading these lines is currenlty in Levi, too!

Been making my own tools for this recently (also rethinking some of the general workflow). So count me in for any discussion. Might need to clean up my code a bit before I can share what I have though 😛

1 Like