search

SciML / StochasticDiffEq.jl

Solvers for stochastic differential equations which connect with the scientific machine learning (SciML) ecosystem
Other
248 stars 66 forks source link

Try bumping RandomNumbers.jl #440

Closed ChrisRackauckas closed 2 years ago

anandijain commented 2 years ago

lol

ChrisRackauckas commented 2 years ago

@frankschae it seems the Strat week are not converging at higher order with the new RNG?

frankschae commented 2 years ago

Hmm... I'll take a look. IIRC most weak solvers are using about 1e9 trajectories for the convergence tests in the respective papers. So, to make them pass for a smaller number of trajectories, it was necessary to tweak them a bit.

ChrisRackauckas commented 2 years ago

A tolerance bump may be required on those stragglers. As long as we know they do converge correctly at 1e9

frankschae commented 2 years ago

yeah maybe I'll double-check 1e9 trajectories outside of the CI.

ChrisRackauckas commented 2 years ago

https://github.com/SciML/StochasticDiffEq.jl/pull/440/checks?check_run_id=3914104494#step:6:363 looks like it is converging, just really really slowly. I'm going to merge since not allowing RandomNumbers.jl is giving the package manager some issues, but this still needs to get sorted out.