debugging speed issue with 2.0

[Bob Carpenter]

Daniel just profiled the time of the log prob call in
2.0 vs. 1.3. The log prob calls are a bit faster now
than they were in 2.0, so the problem's definitely in
the adaptation.

What I'm seeing is that if I stop adaptation after 100
iterations, the result's pretty similar to what we were
getting in 1.3. But if I let 2.0 keep adaptating, then it
really bogs down.

Running through the Instruments profiler in Xcode didn't
show much difference in terms of where the time was being
spent, so I'm really confused about why the iterations are
so much slower in 2.0 than in 1.3.

- Bob

[Allen Riddell]

In the event you folks could use another example, I'm attaching a toy linear
regression example using the horseshoe prior.

1.3.0: 2.5 minutes
2.0.0: 6.8 minutes

Thanks,

-a

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.

[Daniel Lee]

Thanks, Allen.

That does help. Simpler examples are a lot easier to work through.

Daniel