Implement RANDOM_UNIFORM function

[travisfranck]

Vensim's RANDOM_UNIFORM (along with other RANDOM_XYZ functions) have not been implemented yet. SDE should support the most used functions, like RANDOM_UNIFORM, RANDOM_POISSON, and RANDOM_NORMAL. Because they all require a random number generator, the suggestion is to focus on one, RANDOM_UNIFORM, to implement the foundation.

Vensim's RANDOM documentation, including the algorithm choice, is documented here:
https://www.vensim.com/documentation/noise_rng.html

A link to some code for the algorithm is here:
https://github.com/MersenneTwister-Lab/SFMT

I asked Vensim forums about tips and pitfalls. TomF provided a good reply on this thread here
https://www.ventanasystems.co.uk/forum/viewtopic.php?p=32192

The highlight of the thread is that:

Equation sorting is deterministic in Vensim, but I'm not sure we could readily tell you exactly what the rules are.
Another source of variation is likely to be calls during initialization - anything that causes one more or one less will break symmetry.

[travisfranck]

Additional thoughts:

• Someone suggested that reproducing Vensim's random number system is "a fool's errand". They suggested having an exogenous stream of random numbers that drives both Vensim and SDE. That would take the algorithm out of the equation, but maybe not the equation ordering.

Some options for implementation:

1. Try the above exogenous stream of random numbers (external file)
2. Try to Vensim's algorithm
3. Just use C's native rand() function
4. Find a way to capture Vensim's random numbers and save it to a file, which is then read into SDE

[MSDRL]

Any news on RANDOM functions implementation schedule?

[chrispcampbell]

Hi @MSDRL, there is no timeline/roadmap for implementing these functions, and to my knowledge no one has fully contributed implementations of the various RANDOM functions to SDEverywhere. I don't plan to work on this myself since we don't make use of these functions at Climate Interactive, but I am happy to provide pointers and review the code if you or anyone are interested in contributing. There is a related thread in discussion #325 that contains some pointers and tips on contributing.

[travisfranck]

@MSDRL I did look into this and talk with SD tool vendors. My memory of the conversations is that it would be difficult (impossible?) to replicate their exact RNG implementations. That means that, while statistically the same, it might not be possible to have SDE reproduce the same results as the original model if it had a RANDOM() function.

On one hand this negates some SDE's comparison and validation functions ("Did the model convert well?") but it also frees up someone to implement an RNG how best they see fit. To make sure it works, you'd have to run the original (Vensim) model 1000's of times and then the SDE version 1000's of times and compare the statistics.

Vensim's RNG is rather well documented and there they have citations to the papers/algorithm that they use. I don't think I ever found a C implementation in the wild, which would make this pretty easy.