3 ms·
I once used symbolic regression (Eureqa) to generate an approximation to the cumulative distribution. From there you can just draw samples from a uniform distri
by Houshalter 11y ago
I once used symbolic regression (Eureqa) to generate an approximation to the cumulative distribution. From there you can just draw samples from a uniform distribution, and map them onto a Gaussian distribution. Some software uses an approximated CDF, using polynomials I think.
- jacobolus 11y agoHere’s a really neat new approach, Olver & Townsend, “Fast inverse transform sampling in one and two dimensions” http://arxiv.org/pdf/1307.1223v1.pdf http://arxiv.org/pdf/1307.1223v1.pdf Built on top of Chebfun, http://www.chebfun.org http://www.chebfun.org
- im3w1l 11y agoHow much faster was it, and how good was the approximation?
- Houshalter 11y agoThe approximation was pretty decent, and it was only a few math functions. I'll try to redo it and compare it to other methods.