#include <RandSVD.hpp>
Public Member Functions | |
| template<typename InputMatrixType , typename UMatrixType , typename SingularValuesMatrixType , typename VMatrixType > | |
| void | operator() (InputMatrixType &A, int target_rank, UMatrixType &U, SingularValuesMatrixType &SV, VMatrixType &V, rand_svd_params_t params, skylark::base::context_t &context) |
Detailed Description
template<template< typename, typename > class SketchTransform>
struct skylark::nla::randsvd_t< SketchTransform >
Definition at line 36 of file RandSVD.hpp.
Member Function Documentation
| void skylark::nla::randsvd_t< SketchTransform >::operator() | ( | InputMatrixType & | A, |
| int | target_rank, | ||
| UMatrixType & | U, | ||
| SingularValuesMatrixType & | SV, | ||
| VMatrixType & | V, | ||
| rand_svd_params_t | params, | ||
| skylark::base::context_t & | context | ||
| ) | [inline] |
Sanity checks, raise an exception if: i) the target rank is too large for the given input matrix or ii) the number of columns of the sketched matrix either:
- exceeds its width or
- is less than the target rank
Apply sketch transformation on the input matrix
The three steps of the sketched svd approach follow:
- apply sketching
- approximate range of A (find Q)
- SVD
Q = QR(Q)
q steps of subspace iteration
Q = QR(A^T * Q)
SVD of projected A and then project-back left singular vectors
Definition at line 42 of file RandSVD.hpp.
References skylark::sketch::c::CWT, skylark::sketch::c::FJLT, skylark::base::Gemm(), skylark::sketch::c::JLT, min(), skylark::nla::rand_svd_params_t::num_iterations, skylark::nla::rand_svd_params_t::oversampling, skylark::nla::rand_svd_params_t::skip_qr, SKYLARK_THROW_EXCEPTION, and skylark::nla::rand_svd_params_t::transform.
The documentation for this struct was generated from the following file:
- /var/lib/jenkins/jobs/Skylark/workspace/nla/RandSVD.hpp
