UncertaintyQuantification.jl
UncertaintyQuantification.jl copied to clipboard
Option to kill and replace long simulations (SLURM)
For certain applications, the time is takes for a simulation to run depends on where it is sampled. This can lead to long runs, sometimes leading to time-outs of the individual runs, or the overall script. Speaking here mainly about subset simulation algorithms.
One suggestion is that we add an option that if a simulation times-out, we replace it (at random) with an already sampled simulation in the chain ... would this break the ergodicity of the chain?