continuous random noise.

[RobertClay]

adding continuous noise into MINOS for continuous variables. This can be done simply by adding noise from numpy.random functions using some variance (e.g. 10 for SF12).

However is seems a good idea to implement common random numbers here to seed values if necessary. vivarium has CRN seeded random number generation but its only for uniform distributed variables (as far as I know https://vivarium.readthedocs.io/en/latest/api_reference/framework/randomness/stream.html#vivarium.framework.randomness.stream.RandomnessStream.get_draw).

You can use two samples from uniform distributions to create two standard Normal N(0, 1) distributions using the Box-Muller transform https://rh8liuqy.github.io/Box_Muller_Algorithm.html. Its pretty straightforward with a couple things to think about.

• [ ] write a box muller transform that draws two uniform distributions in python and outputs two standard normal distributions. We only need one and will have the other spare. Some way to utilise this somehow? Store it in a buffer somewhere. Not very computationally expensive so not going to bother for now.
• [ ] scale BM variable according to desired variance. Need to times variables by some standard deviation value sigma (NOT sigma^2).
• [x] add to updated variable (e.g. SF12) before submitting to population manager.
• [x] maybe a more generalised CDF inversion method? Probably quicker and only generates ones vector. generate uniform rv, put it through PHI, scale by sigma.

[RobertClay]

well that was easy.

[RobertClay]

very easy to add Gaussian noise on top of continuous variables. Using scipy.special.ndti can convert uniform distribution draws to Gaussian variables. then scale by desired mean and sd. Couple of follow up questions:

• What is the right amount of noise to add? Just adding noise seems unhelpful. Filtering methods may be useful here.
• Non-Gaussian noise? SF12 pretty heavily skewed. Is there better noise than can be added? e.g. Gamma noise.