Closed spinicrus closed 4 years ago
Hi spinicrus,
the problem I think the problem is that your (40k, 500) matrix would be decomposed into Q (40k, 40k) and R (40k, 500) and Q would consume about 12GB of memory (40k40k8), and you'd also have a copy to call the native side of things. What do you want to achieve? Maybe if you'd do a QR decomposition on the transpose, would that help?
-M
yes you are right M, thank you very much! i need it in a method that does merging of two simple lsa models (for some text analysis), and for that i would need q&r. but maybe i can dance a little :) although as a quick temp resolve i used mtj just for this particular task.
Hi, thanks for jblas!
got stuck doing qr decomp on a 40k:500 DoubleMatrix. -Xmx9g, but it doesn't reach it, somewhere at 2.5g+ it happens. Any advice would be appreciated, thanks a lot. (maven installed on os x 10.11.6 (el capitan) i5)