"it is always safe to assume dependence" - to_event vs plate

let me rephrase. if you ask something like o3 targeted questions and make sure it justifies it answers with references to pyro code/documentation, you might get some good info. i wouldn’t rely on it to make complex summaries though. this summary is crude and not really accurate/helpful.

what i’ve been trying to tell you is that there is no simple table like this. there are too many subtle edge cases. what pyro does under the hood depends on a myriad of details. what is the dependency structure of your model? do you have discrete latent variables? is there subsampling (not generically supported in mcmc)? are you using variational inference, mcmc, or something else? etc etc etc

give pyro as much information about your assumed model structure as possible by using plates everywhere and, failing bugs, pyro should do the right thing even if you don’t understand what’s going on under the hood. if you do want to understand what’s going on under the hood, you have to grok relevant references.

Thanks again for your time, Martin ! Got it, I have not yet learned how to use ChatGPT well enough for this purpose. I hear you that there is no such simple table. I really am trying just to understand what is safe about assuming dependence, because your statements below seem to contradict “it is always safe to assume dependence” ? It may not be safe if folks (like me) don’t know all the things under the hood ?

give pyro as much information about your assumed model structure as possible by using plates everywhere and, failing bugs, pyro should do the right thing even if you don’t understand what’s going on under the hood. if you do want to understand what’s going on under the hood, you have to grok relevant references.

safe here means “mathematically permissible”