RNG isolation

[michael...@gmail.com]

Greetings all...

I'm working on some discrete simulation prototypes and have a question regarding the isolation of random number generation states.

From looking at `distribution.js` and `jstat.js`, most calls use the built-in `Math.random` function. While this can be overridden at the global level (yuck!) with another RNG (to support seeding, etc.), it still refers to a singular RNG, and assumes it's the same RNG for all use cases.

However, I'm wondering if any work/thought has been given to partitioning/isolating RNG state? In the simulation case, I have several input variables that each need to have an isolated random number stream generated regardless of how many other variables may have been sampled.

For example, assume 2 input variables 'X' (Normal) and 'Y' (Poisson). If I repeatedly alternate between sampling 'X' and 'Y' over 100 samples, I get a specific stream of numbers for each variable. The problem is that today, if I remove 'Y', the number stream generated for 'X' is no longer the same (due to the internal state of the RNG not being mutated for calls to 'Y').

Any pointers / thoughts / comments welcome.