All, I've been working on another implementation of SVD as in issue MATH-355<https://issues.apache.org/jira/browse/MATH-355>, I'm mostly there. However, I'm consistently impressed with the JAMA code. I email to ask a question that is perhaps quite bold for one such as me, so new to Commons Math.
I'm using both the existing SVD implementations in CM and in JAMA as benchmarks. However, JAMA code is hard to beat, it's fast and stable. Additionally, current issues such as MATH-383<https://issues.apache.org/jira/browse/MATH-383>, MATH-465 <https://issues.apache.org/jira/browse/MATH-465> an MATH-583<https://issues.apache.org/jira/browse/MATH-583> are not suffered by JAMA. JAMA is not only faster and more accurate for SVD, but also for QR-decomposition. There may be other examples where JAMA is good, I'm yet to look. CM is a great package, but I email to inquire if could we could solve easily the issues above by simply implementing public-domain JAMA-like code within the linear algebra sub-package or, perhaps more controversially, have JAMA as a dependency to CM? Is 'home-grown' code over public-domain code an objective of Commons Math? Like I say, it's a bold question. Anticipating, Chris
