A stupid name for a smart matrix library, because who doesn't love smart matrices? Being implemented as a data type around the native BLAS hooks of jblas gives it speed. Being implemented as a Clojure sequence makes it clever.
For now, you can read the Marginalia documentation or take a look at a few examples below. (Note: To be updated!)
(in-ns 'clatrix.core)
(matrix (repeat 5 (range 10)))
; A 5x10 matrix
; -------------
; 0.00e+00 1.00e+00 2.00e+00 . 7.00e+00 8.00e+00 9.00e+00
; 0.00e+00 1.00e+00 2.00e+00 . 7.00e+00 8.00e+00 9.00e+00
; 0.00e+00 1.00e+00 2.00e+00 . 7.00e+00 8.00e+00 9.00e+00
; 0.00e+00 1.00e+00 2.00e+00 . 7.00e+00 8.00e+00 9.00e+00
; 0.00e+00 1.00e+00 2.00e+00 . 7.00e+00 8.00e+00 9.00e+00
(from-indices 5 5 *)
; A 5x5 matrix
; ------------
; 0.00e+00 0.00e+00 0.00e+00 0.00e+00 0.00e+00
; 0.00e+00 1.00e+00 2.00e+00 3.00e+00 4.00e+00
; 0.00e+00 2.00e+00 4.00e+00 6.00e+00 8.00e+00
; 0.00e+00 3.00e+00 6.00e+00 9.00e+00 1.20e+01
; 0.00e+00 4.00e+00 8.00e+00 1.20e+01 1.60e+01
(time (solve (rand 1000 1000) (rand 1000)))
; "Elapsed time: 158.216 msecs"
; ....
(time (rank (rand 1000 1000)))
; "Elapsed time: 13469.51 msecs"
; 1000
(let [A (rand 10 14)
B (* A (t A)) ; B is symmetric
lu (lu B)
P (rspectral 10) ; `respectral` makes positive definite matrices
G (cholesky P)] ; so we can get their square root
(and (= A (c/t (c/t A)))
(= B (* (:p lu) (:l lu) (:u lu)))
(= P (* (t G) G))))
; trueCopyright © 2012 Joseph Abrahamson and contributors (see LICENSE.txt)