This package provides an implementation of the FameSVD algorithm via the BLAS and LAPACK routines syrk
, syevr
and gemm
, The provided method is faster than the SVD algorithm used in the Julia standard library and as shown in the paper faster than the Krylov-Method and Randomized-PCA.
Please note that column size was kept contstant at 1000 and the machine used had 16GB DDR4 RAM and an Intel i7-8565U CPU running at 4.6GHz.
The package provides the function fsvd
which returns an LinearAlgebra.SVD
object.
S = FameSVD.fsvd(A)
Xiaocan Li, Shuo Wang and Yinghao Cai: "FameSVD: Fast and Memory-efficient Singular Value Decomposition"; arXiv:1906.12085v1