Daniel Lee

That sounds great! We just need to deal with the platform dependence, but maybe we can work through that by storing different outputs. > On Jun 17, 2017, at 11:23...

This would be in addition to the statistical correctness tests. If we are serious about having more contributors, we have to make sure things don't inadvertently change. For example, if...

My main concern is with the stability of our code base. If I can't trust that code runs the same merging with develop, then it gets harder and harder to...

If there's any other / easier way of accomplishing it, please suggest it. I don't think golden tests are ideal, but if there isn't a viable alternative, I'd rather have...

I hear the arguments for it being a burden. But I think we're framing it the wrong way. It should be seen as something to make sure the rest of...

> On Jun 19, 2017, at 8:43 AM, Michael Betancourt wrote: > > Your description is far too subjective and illustrates the asymmetric > nature of the burden of these...

There's one more relevant test case: - changes to the code base that pass unit tests, but change behavior in the samplers due to some side-effects. Gold tests would also...

@bob-carpenter: good points. Fortunately, we should be able to test with just the Stan library using callbacks. We could also test with the interfaces. Agrees that it's going to be...

Great list. I don't think there's any harm in introducing this earlier than an actual Stan v3 release if any one has time to tackle it.

Yup! You're absolutely right. I'll get to it soon. On Tue, Feb 6, 2024 at 11:01 AM Steve Bronder ***@***.***> wrote: > I think we need an opencl version of...