Closed epiben closed 1 year ago
Implemented in da4002b. Now, the user can specify the maximum number of participants to use per batch, ensuring that batch sizes are be the same regardless of trial sizes. This should facilitate estimation of memory use so as to unify resource needs (at least w.r.t. RAM) across scenarios, in turn making it easier to exploit workload managers such as SLURM when many scenarios are run in parallel.
The current implementation doesn't distinguish between trials with many and with few participants even though this is crucial: simulating 50k trials with 200 participants (= 10M participants in total) might be feasible in one go, while batching is likely required with e.g. 4000 participants per trial (= 200M participants in total).