Closed alecksphillips closed 1 year ago
I'm conscious that this approach goes against keeping lazy-evaluation in the rear view mirror, but me and @AVarsi88 figured this would be necessary for reproducibility on distributed architectures.
Thanks Nav.
Move the RNG to a single RNG class. For SMC, each particle should be allocated an instances of this RNG class. This way, running SMC in parallel with a given number of particles should be reproducible independent of the number of cores/nodes used to do the sampling.