Open 1fish2 opened 6 years ago
My 2 cents: From a theoretical perspective, model outcome should be the same regardless of process execution order, since those processes should be operating in parallel. This would require each process to have its own random number generator, which was the case in the past, but based on a conversation with Derek a while back, I believe this was changed to a single random number generator. Practically speaking, this should be fine, so long as execution order is always the same.
Regardless, this should be tested in the following fashions:
... to ensure that the psuedo-random sequences are computed consistently, if feasible.