interested to hear if anyone is using #stan @mcmc_stan in industrial settings? Can be based in industry or academic collaboration
@_jcken @mcmc_stan It's pretty popular in the pharmaceutical industry (although e.g. jags is also popular), including for using priors based on historical data, implementing flexible multiple imputation, getting uncertainty when there's no obvious frequentist to do so, good small sample performance etc.
@_jcken @mcmc_stan
And of course the workflow of using a Bayesian posterior as an input to a decision analysis is very well suited to decision making situations. Personally, I try to go via brms as a front-end (a lot less time spent writing code when it suits the problem).
@BjoernHolzhauer I can believe rstanarm/brms/prophet etc are used more often since easier to work with! Where I work, we offer stan training courses and it's almost always academics who come, so was interested to see if there's a need elsewhere
@_jcken there's definitely some Stan usage via ratan/cmdstanr. But to be fair what most people need most of the time is covered by higher level interfaces to Stan. It's actually kind of hard to write more efficient code than, say, brms, plus it's just faster... Nevertheless, occasionally, I really do write Stan code.
@_jcken @mcmc_stan There are plenty, at least based on the affiliations of those who go to Stancon's (e.g. in StanCon Helsinki 30+ companies)
@_jcken @mcmc_stan while it doesn't use mcmc by default, the Prophet library uses it behind the scenes. That was a bit of a hype a few years ago as the new cool timeseries analysis tool.
@_jcken @mcmc_stan There seems to be a lot of use of Stan in the industry. The first two examples that come to my mind are Pharmacometrics ---we have some examples in Section 1.1 of this paper (https://ascpt.onlinelibrary.wiley.com/doi/10.1002/psp4.12812); and forecasting in tech, notably via the software Prophet (https://facebook.github.io/prophet/) built on top of Stan.