Random number generation for non-uniform distributions

[stress-tess]

There has been user desire for generating random numbers with respect to other distributions (normal in particular). But we should also consider adding other common distributions (power law, binomial, poisson, etc)

Right now we use fillRandom in chpl which doesn't seem to have the ability to specify distributions. I know there are ways of converting uniform into other distributions and bill did a power law conversion for the sort benchmarks. I'd rather not get too custom with it though since random number generation is very easy to mess up

I think we need to dive into how numpy does randomness to try and align ourselves with them as much as possible

• create random subfolder and foundation for a generator #2993
• Add generator sym entry and stateful uniform distribution  #3008
• Add documentation for our random number generation #3017
• Add choice, permutation, and shuffle to random number generators #3066
• Add normal to random number generators #3167
• Add exponential and standard_exponential to random number generators #3183
• add logistic to random number generators #3372
• add lognormal to random number generators #3373
• add multivariate_normal to random number generators #3374
• add poisson to random number generators #3245
• add gamma and standard_gamma to random number generators #3846

here is the full list of functions supported by numpy's generators that have yet to be implemented, categorized according to how important they seem to me (i.e. do I recognize it from the single stats class I've taken?)

remaining np random methods

Prioritized:

• numpy.random.Generator.beta
• numpy.random.Generator.binomial
• numpy.random.Generator.chisquare
• numpy.random.Generator.laplace
• numpy.random.Generator.geometric
• numpy.random.Generator.logseries
• numpy.random.Generator.multinomial
• numpy.random.Generator.power

Not prioritized at the moment:

• numpy.random.Generator.permuted
• numpy.random.Generator.dirichlet
• numpy.random.Generator.f
• numpy.random.Generator.bit_generator
• numpy.random.Generator.spawn
• numpy.random.Generator.bytes
• numpy.random.Generator.gumbel
• numpy.random.Generator.hypergeometric
• numpy.random.Generator.multivariate_hypergeometric
• numpy.random.Generator.negative_binomial
• numpy.random.Generator.noncentral_chisquare
• numpy.random.Generator.noncentral_f
• numpy.random.Generator.pareto
• numpy.random.Generator.rayleigh
• numpy.random.Generator.standard_cauchy
• numpy.random.Generator.standard_t
• numpy.random.Generator.triangular
• numpy.random.Generator.vonmises
• numpy.random.Generator.wald
• numpy.random.Generator.weibull
• numpy.random.Generator.zipf

[stress-tess]

I found the c code where numpy defines it's distributions. It uses pcg as it's random number generators, same as chapel. So it feels like we could do something similar, but there's more digging to be done to tell how well this ports to chpl