Ralf Stubner

For now it looks like adding the same specialization is the way to go. In addition, I should also investigate to what extend these specialization are actually needed, i.e. if...

Can you provide an explicit example? Vicente J. Ontiveros ***@***.***> schrieb am Mi. 2. Aug. 2023 um 15:55: > I just made a microbenchmark of sample.int versus dqsample.int and while...

To be more explicit, I ran a quick test myself: ``` r library(dqrng) m 1 sample.int(m, n, replace = TRUE, prob = prob) 13.24ms 13.95ms 70.8 #> 2 dqsample.int(m, n,...

Thanks! I have taken the liberty to change the formatting of your code by surrounding it with three backticks. This way it is better readable. Using your code (and using...

The timing differences for the R functions are noticeable but not that extreme. I would not get a new computer just now. The differences for `dqrng` are extreme, though. Maybe...

You are welcome. Actually I would like to keep the issue open since I did see some strange effects. For a population of 10 drawing one 1 item `dqsample.int` is...

Unfortunately these tests are only scratching the surface. The performance from probabilistic sampling gets really bad, if one (or few) possibilities have much higher weights than the others. To quantify...

Thanks for the suggestion! I am still interested in adding `sample` functionality to `dqrng`. However, I have no immediate need for it, so I cannot give a time frame for...

BTW, in the original paper the timings look much better for the StochasticAcceptance algorithm, being faster than BisectionSearch even without the (IMO) artificial change in weights. I guess that for...

I am finally working on this: https://stubner.me/2022/12/roulette-wheel-selection-for-dqrng/, looks promising.