la4j
la4j copied to clipboard
SVD should work fine with [m < n] matrices
We need to handle this.The current implementation expects to see [m >= n] input matrix.
Could you give me some clues about SVD? What SVD is and where can I find it?
This is very tricky issue. The current implemention is derived from JAMA. So, initial JAMA authors (MathWorks fellows and principal engineers) didn't implement such functionallity. Which means, that there are some pitfalls with it.
This is a long-term issue. So, I would suggest you to look at this issue: https://github.com/vkostyukov/la4j/issues/135. It should be easy to implement it. But we have to understand how to design it.
I think we should implement SVD from scratch. This is a good paper to start with: http://www.cs.utexas.edu/users/inderjit/public_papers/HLA_SVD.pdf