Stanc3 optimization and analyses walkthrough during StanCon

Hey all,

@enetsee wants to talk about the optimization and analyses passes in stanc3 with @Matthijs sometime during StanCon, hoping to get a walk through of what the code is intending to do (especially for lazy code motion). I’m also interested in attending if I can make it. Anyone else interested in this topic?

-Sean

7 Likes

I’m in, obviously!

Me and @tadej are also interested, even if its probably above our “paygrade”. The issue is that we will only be at StanCon on Thursday & Friday. But no worries if others would rather schedule it on the tutorial days.

1 Like

Will this material be made available online / elsewhere?

I’m very interested, but wont be at StanCon.

I have been slowly working on ProbabilityModels.jl, a Julia library that provides a DSL for fitting with NUTS. It is light on features, and basic control flow like if/else and loops are a ways away (but I plan on at least supporting broadcasting soon), but the focus has been on high performance.
The example in the README sampled something like 20 times the effective samples per second, so I am interested in all the optimization work going on.

1 Like

There’s no optimization you’d be seeing yet, haha. You may be more interested in the Stan automatic differentiation paper for looking into why Stan is so fast.