Matthijs Vákár
Matthijs Vákár
This is difficult to add to Stan currently as it would lead to ambiguity. For example, is the following a valid Stan program if we add this behaviour ``` model...
This is difficult to add to Stan currently as it would lead to ambiguity. For example, is the following a valid Stan program if we add this behaviour ``` model...
Is this still desired, or should we close the issue?
Does `log_mix` address this issue?
Ah, I guess it would also depend on the design in the surface language, right? What are you planning to do in that respect? I'd always had a design in...
(Sorry if what I'm saying is naive. I've been out of the loop for a year, so have no idea what the current state is of GPU support.)
Very cool! (Sorry, just seeing this now. I need to make sure github emails end up in my inbox.) I very much agree that the strength of Stan is that...
Also, those speed-ups are insane! So cool! This'll allow Baysian inference to scale to totally new applications once it's done.
Even with this design where the compiler essentially decides what gets sent to GPU, perhaps it's useful to just have explicit types for GPU-matrices etc in the MIR? They wouldn't...
> There are similar figures to the binomial one for almost all lpdf/lpmf functions: https://unilj-my.sharepoint.com/:f:/g/personal/rokcesnovar_fri1_uni-lj_si/EgPP9IsCdFJJg0Ygm4_4-HcB9FOozC3FayacTGXeaT6LEg?e=ULRnwA > > Not sure anyone is interested but posting anyway :) Of course people are...