search

SciML / JumpProcesses.jl

Build and simulate jump equations like Gillespie simulations and jump diffusions with constant and state-dependent rates and mix with differential equations and scientific machine learning (SciML)
https://docs.sciml.ai/JumpProcesses/stable/
Other
135 stars 35 forks source link

Allocations test fix #328

Closed isaacsas closed 1 year ago

isaacsas commented 1 year ago

It seems that it currently takes a couple runs of RSSACR before the table allocations are fully settled down now. It is expected that there can be some allocation differences here for the algorithms that use PriorityTables (which can dynamically allocate additional buckets, but only ever grow, so should settled down eventually).

isaacsas commented 1 year ago

My only confusion is to why this issue has cropped up now all of a sudden.

ChrisRackauckas commented 12 months ago

This popped up when v1.9 came out.

isaacsas commented 12 months ago

Huh, I never saw that issue before (even with the previous PR tests passed).

In any case I did dig into it with the allocations profiler and it seemed to be coming from additional push!s into groups within the priority table as best I could tell (and not the function wrapper type instability issue we had before). Moving this test to use StableRNGs should hopefully eliminate the issue.