@SGB At today’s general meeting (April 2, 2020) Andrew brought up the reputation of Stan as not being fast in comparison to various competing implementations. It was noted that this is often due to badly written models. I’ll start with my contribution but will list what I recall from others as well…

[image] ChrisRackauckas: I think the main issue with Stan in this department is really just that the kernels are all baked in. Seeing an FFI ( Foreign Function Interface in Stan’s future ) would be nice, because then external developers like us could add new differential equation solvers to Stan b…

Sorry, I did not mean to imply that there is not principled modeling outside of Stan. Speed of course matters, but my experience is similar to those above that parameterization almost always gains more speed than facilities. Of course, they work together and coordination of the guts of these things …

Just wanted to say I appreciate everyone taking the time to chime in here. Thanks to the Stan users and devs as well as the Julia and Turing folks!

[image] grburgess: One can certainly autogrid through all of Julia, but does it translate to stability for free? I remember being frustrated that Stan couldn’t do high-D interpolation out of the box… but you begin to understand that this is a very bad thing to do with HMC in the first place. In …

We’ve been using Stan for DDE and SDE problems for years now, and the difficulty is always parametrizing the model for good posterior geometry. This is quite the same point that you are making in your own SciML and UDE papers, namely that better model structure is the way forward for machine learni…

[image] jonah: Just wanted to say I appreciate everyone taking the time to chime in here I’ll second that. Sorry I missed the last month of conversation somehow! [image] mohamed82008: This reply turned out longer than I had liked. To the admins, feel free to move this comment to a separ…

[image] Bob_Carpenter: What’s a CFD simulation? Computational Fluid Dynamics. What I’ve done for PDE is a proof of concept, more realistic work has been started by @IvanYashchuk in his petsc project Using PETSc library together with Stan . I don’t think it’s meaningful to jump into large-scale…

[image] Bob_Carpenter: That’s 10,000 Julia developers over all? How many write the kind of differentiable code you can use in Turing.jl? I really like the basic idea and understand the argument for it. My understanding is that it’s implemented something like the way “object oriented” code is im…

[image] Bob_Carpenter: My advice is to do what we did for our autodiff paper on arXiv—write to all the developers of those other systems with your benchmark code and ask them if there’s a way to improve it before publishing misleading benchmarks. It’s a lot of work, which is why we usually don’t…

Also I won’t speak for @mans_magnusson but I’d really like the idea of reaching out to other PPL developers and pitching posteriorDB as a good solution for testing across languages. idk how ready it is for that but if we want something everyone uses then I think it would be good to get feedback from…

Addressing Stan speed claims in general

Stan Governance

martinmodrak Split this topic January 15, 2021, 7:20pm 81

7 posts were split to a new topic: Stan backend for NumPyro + performance comparison

Stan backend for NumPyro + performance comparison

Topic		Replies	Views
Any way to make Stan competitive with Tensorflow for maximum likelihood? Algorithms	43	6598	July 31, 2019
"Selling" Stan General	14	5461	April 7, 2018
"Wiki" or "Facebook" for Stan models; building the Stan community Developers	29	2805	November 19, 2016
Modularity for making Stan more Pythonic and Rthonic Developers	54	3428	October 9, 2019
Stan backend for NumPyro + performance comparison Publicity	10	3455	January 17, 2021

Addressing Stan speed claims in general

Related topics