Know if the model is having problems before it finishes


#1

I’m not really sure where to raise this issue. But I’m fitting a very problematic model, I can already tell that it’s going to fail if one of the chains is much slower than the others, but it would be great to know this for sure while the model is running (and waste less time).

It would be great to get if there are divergent transitions while the model is sampling, and/or some mean Rhat based on the current samples and all the parameters Rhats.

Bests,
Bruno


#2

the R package ctsem includes a function called ‘stanWplot’ that is just a wrapper to the stan function which loads a browser window and shiny app for plotting the current sampling state. So instead of calling stan just call ctsem::stanWplot

But, if this is just a general suggestion, I agree! :)


#3

You mean stanWplot?
So it works for any stan model right, not only the time series models that you do with your package, right? Very, cool! But I’m working on a server so, the browser will be problematic :(


#4

right don’t even remember my own functions, and yeah it’s a general stan function :) I see re the server problem. The code shouldn’t be too hard to adapt to run on a workstation and point to the relevant folder.