SciML / StochasticDiffEq.jl

Solvers for stochastic differential equations which connect with the scientific machine learning (SciML) ecosystem
Other
251 stars 67 forks source link

WeakConvergence tests are not reliable #386

Open ChrisRackauckas opened 3 years ago

ChrisRackauckas commented 3 years ago

A bit of stochasticity can make the convergence estimate fail, a la: https://buildkite.com/julialang/stochasticdiffeq-dot-jl/builds/40#52ddc367-5e7c-4deb-a320-193cf981b835/359-869 .

@frankschae has played with this a ton, and it seems like it is a test that just needs more compute power, more trajectories, and more parallelism. This might be a candidate to use GPUs on or AMDCI.