Most PRNG functions in R rely on random unif/norm/exp numbers, e.g., rbeta uses a rejection method relying on uniform random numbers (github.com/wch/r-source...) - if they were reimplemented with the RNG calls replaced with the ziggurat functions, I would imagine that all could be speed up?

Are there any plans to expand to cover PRNG of other distributions in R? Would be really useful (I use rbeta a lot). A lot of them seem to rely on random uniform/normal distributions, so I would assume they could be sped up as well if the ziggurat algorithm was used under the hood? 2/2

Very interesting work! I use a lot of PRNG for simulation, and it often takes the majority of computation time (50-80%). On this machine the speedups seem around times 6 (norm), 8 (exp), but somewhat less for unif (~1.9) 1/2

Feedback, comments, and questions on the preprint is welcome! 4/4

This framework is used for our upcoming EMPRESS trial (www.cric.nu/empress) and all currently planned domains on our upcoming INCEPT platform trial (incept.dk).

We hope that this will serve as a useful help for others considering or planning the use of similar trials.
3/..

We cover Bayesian adaptive trials with adaptive stopping and arm dropping and response-adaptive randomisation. In addition to describing and providing guidance on the methodological decisions, we included complete analysis code to evaluate such trial designs.
2/..