Skip to content


Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
branch: master
Fetching contributors…

Cannot retrieve contributors at this time

81 lines (72 sloc) 2.433 kb
#lang scheme/base
(require scheme/math
(planet williams/science:3/random-distributions/gaussian)
(define 2pi (* 2 pi))
;; (vectorof number) (triangular-matrixof number) -> (vectorof number)
;; mu is the mean
;; sigma is the square root of the covariance matrix -- use
;; the cholesky decomposition to calculate this
;; Uses the method given at Wikipedia
(define (random-multivariate-gaussian mu sigma)
(define n (vector-length mu))
(define z
(for/vector ([i n])
(random-gaussian 1 0)))
(vector+ mu (vector*m z sigma)))
;; (vectorof number) (vectorof number) (matrixof number) -> [0,1]
;; x is the point
;; mu is the mean
;; sigma is the cholesky decomposition of the covariance
;; matrix, as created by matrix-cholesky (lower triangle is
;; L, upper is L^T)
(define (multivariate-gaussian-pdf x mu sigma)
(define n (vector-length mu))
(define det (matrix-cholesky-determinant sigma))
(define inverse (matrix-cholesky-invert sigma))
;; Normalisation constant
(define z (/ 1 (* (expt 2pi (/ n 2)) (sqrt det))))
(define diff (vector- x mu))
(define p (exp (* -1/2 (vector-dot (vector*m diff inverse) diff))))
(* z p))
;; (vectorof number) (matrixof number) ->
;; (values (-> (vectorof number)) ((vectorof number) -> number))
;; mu is the mean
;; sigma is the covariance matrix
;; Returns two functions, the first generates samples, and
;; the second is the PDF. This is more efficient than
;; repeatedly performing the required matrix operations.
(define (make-multivariate-gaussian mu sigma)
(define n (vector-length mu))
;; Cholesky is faster but LU is more numerically stable
;; AND valid for singular matrices
(define cholesky (matrix-cholesky sigma))
;;(define det (matrix-cholesky-determinant cholesky))
;;(define inverse (matrix-cholesky-invert cholesky))
(define-values (lu p s) (matrix-lu sigma))
(define det (matrix-lu-determinant lu s))
(define inverse (matrix-lu-invert lu p))
;; Normalisation constant
(define z (/ 1 (* (expt 2pi (/ n 2)) (sqrt det))))
;; Sampler
(lambda ()
(random-multivariate-gaussian mu cholesky))
;; PDF
(lambda (x)
(define diff (vector- x mu))
(* z (exp (* -1/2 (vector-dot (vector*m diff inverse) diff)))))))
Jump to Line
Something went wrong with that request. Please try again.