Advanced statistics routines
56 views
Skip to first unread message
Brendan Tracey
Jun 14, 2016, 1:00:15 PM6/14/16
to gonum-dev
I have been using advanced Monte Carlo integration routines recently, specifically the algorithm in "Control Functionals for Monte Carlo integration" https://arxiv.org/pdf/1410.2392v5.pdf and in my own work on "Stacked Monte Carlo" https://arxiv.org/abs/1606.02261 .
In naive Monte Carlo, one takes the average of the function samples. It turns out you can get drastically reduced error by doing smarter things (orders of magnitude better).
It would be nice to provide these capabilities to users of Go, but I'm not sure where they belong. If they belong in gonum, it seems like they would belong in intergrate/mc. There is definitely a fuzzy area between that and stat/sample, though I would say stat/sample is just about generating values from probability distributions, while integrate/mc is about finding expectations of functions under distributions. Also in intergrate/mc (in my mind) would go adaptive sampling routines that try to select good points based on low error.
If they don't belong in gonum, is there somewhere that they do belong?
In naive Monte Carlo, one takes the average of the function samples. It turns out you can get drastically reduced error by doing smarter things (orders of magnitude better).
It would be nice to provide these capabilities to users of Go, but I'm not sure where they belong. If they belong in gonum, it seems like they would belong in intergrate/mc. There is definitely a fuzzy area between that and stat/sample, though I would say stat/sample is just about generating values from probability distributions, while integrate/mc is about finding expectations of functions under distributions. Also in intergrate/mc (in my mind) would go adaptive sampling routines that try to select good points based on low error.
If they don't belong in gonum, is there somewhere that they do belong?
Seb Binet
Jun 15, 2016, 5:33:22 AM6/15/16
to Brendan Tracey, gonum-dev
gonum/integrate/mc sounds good.
-s
Reply all
Reply to author
Forward
0 new messages