SciMLBenchmarks.jl
SciMLBenchmarks.jl copied to clipboard
SciML
Jump benchmarks
Resurrecting and updating https://github.com/SciML/SciMLBenchmarks.jl/pull/298
Is it ok to just edit it to 1.6, or do I need to regenerate it under 1.6 to make sure it is consistent / works ok?
@Vilin97 I think I've got this fully updated now, and it looks ok locally, but I do wonder if there is an issue in DirectCRDirect. It is still showing the same asymptotic scaling as NSM, and not matching up with Sanft's results where the work plateaus. I added a callback to make it actually stop simulating at 1e8 events, so it should be replicating what he did. I wonder if we need to go to a larger problem size to see the correct asymptotic scaling, though it seems weird we are well-past where he saw it and it hasn't kicked in...
I giving this a last run locally to check the html output, and will then need to regenerate the Manifest on 1.7 which I don't currently have installed, but otherwise this is hopefully all good finally.
I'll update the remaining jump benchmarks to work with this manifest/project and Catalyst.
@Vilin97 I think this is done -- let me know once you've taken a look at the spatial benchmark and then we can merge if tests have passed.
@Vilin97 thanks for taking a look! For future reference, feel free to use the code suggestion mechanism for small changes like removing the redundancies. Then I can just commit them directly from your comment without having to edit anything locally.
@ChrisRackauckas I think this is good to go now. I am running the spatial benchmark today locally if you want to wait until I get confirmation it finishes properly (I have run it up to N=120 mesh voxels, but not at the largest size in the PR, 240).
Yeah I've been following this. It looks good. There is still a depwarn in there but I'm not sure where that's coming from.
The package version issue doesn't come up locally for me. Could it be because I'm running on a 1.8 rc? Maybe that causes issues with the manifest (though the warning is about packages loaded from the SciMLBenchmarks project file)?
This issue
The variable syntax (G[1:2N])(t) is deprecated. Use (G(t))[1:2N] instead.
is coming from the problem library I think, which has not yet been updated for this depreciation (since it is a recent change in Symbolics).
This one
┌ Warning: Weave: `options` key is deprecated. Use `weave_options` key instead.
seems to be from this PR -- I can update the jmd files with the new notation when I get to work a bit later this morning.
It’s finished. Tests just failed from the server job timeout being too short I believe.
Yes, 16h and 40minutes is 1000 minutes which is the time out time. Let's merge this so most of them are updated, and then just update the long one separately. I think on its own it doesn't hit the limit, but all jump benchmarks together does hit the limit.