Implement generation of correlated samples


(Philip G) #1

Hi,

I’m a big fan of the Math.net library. Especially the distribution section has saved me a lot time, which would have been spent implementing stuff on my own.

But one big thing missing for me is the ability to generate correlated samples, i.e. create samples using copulas. In my area of stochastic Monte-Carlo simulation it is only scarcely that independent random samples are needed. Most models are built on correlated variables.

I would like to implement a few methods that allow sampling from the most common copulas (Gauss, t - copula, some Archimedean ones).
As far as I know there is no open source library for .net available right now, which offers this functionality.

If anyone has any ideas / requests or reasons why this is not a good idea for Math.net please contact me, otherwise I will start a fork of the library and open a pull request when the foundation is laid down.


(Christoph Rüegg) #2

It seems to me this would fit very well to the scope of Numerics. I’d certainly accept contributions in this area!


(Philip G) #3

For anyone interested, I’ve begun implementing some Copulas.
You can follow the project on https://github.com/PGsHub/mathnet-numerics
I’m open to any input so please just contact me anytime. Since I am doing this mostly on the weekends and late in the evening during the week, the progress is not too fast. But I hope to have enough implemented soon, to think about merging it into the mein library.