Hi,

I’m wondering if there is a way of generating singular value sequentialy. I’m aware of the truncated svd utilities from TSVD.jl and LowRankApprox.jl but to my knowledge they didn’t provide a way to generate singular value sequentialy. I mean if I have the r first singular value and want the r + 1, I will have to recompute the r first…

If you have the first r singular vectors as well as the singular values, then you can deflate the matrix — project onto the orthogonal complement of the singular vectors — in order to remove those singular values.

3 Likes

Thank you for your answer, yes it is a solution !

I was thinking of a way to build a generator function (like with the yield syntaxe in python) based on the existing tsvd function but it’s definitely easier to simply remove the r first singular value as you suggest, although certainly less efficient !