diff options
Diffstat (limited to 'unsupported/Eigen/src/BDCSVD/TODOBdcsvd.txt')
-rw-r--r-- | unsupported/Eigen/src/BDCSVD/TODOBdcsvd.txt | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/unsupported/Eigen/src/BDCSVD/TODOBdcsvd.txt b/unsupported/Eigen/src/BDCSVD/TODOBdcsvd.txt new file mode 100644 index 000000000..0bc9a46e6 --- /dev/null +++ b/unsupported/Eigen/src/BDCSVD/TODOBdcsvd.txt @@ -0,0 +1,29 @@ +TO DO LIST + + + +(optional optimization) - do all the allocations in the allocate part + - support static matrices + - return a error at compilation time when using integer matrices (int, long, std::complex<int>, ...) + +to finish the algorithm : + -implement the last part of the algorithm as described on the reference paper. + You may find more information on that part on this paper + + -to replace the call to JacobiSVD at the end of the divide algorithm, just after the call to + deflation. + +(suggested step by step resolution) + 0) comment the call to Jacobi in the last part of the divide method and everything right after + until the end of the method. What is commented can be a guideline to steps 3) 4) and 6) + 1) solve the secular equation (Characteristic equation) on the values that are not null (zi!=0 and di!=0), after the deflation + wich should be uncommented in the divide method + 2) remember the values of the singular values that are already computed (zi=0) + 3) assign the singular values found in m_computed at the right places (with the ones found in step 2) ) + in decreasing order + 4) set the firstcol to zero (except the first element) in m_computed + 5) compute all the singular vectors when CompV is set to true and only the left vectors when + CompV is set to false + 6) multiply naiveU and naiveV to the right by the matrices found, only naiveU when CompV is set to + false, /!\ if CompU is false NaiveU has only 2 rows + 7) delete everything commented in step 0) |