Andreas Borgen Longva

Thanks for bringing up the topic. Let's make one thing clear first. Are we talking about interop for *arrays* (potentially multi-dimensional of many dimensions) or *matrices* (and possibly vectors)? Or...

Sorry for not keeping up with the conversation, I'm a little busy these days. I think perhaps this is evolving into something more complicated than it needs to be, and...

Note that I was referring explicitly to the absolute difference test for equality in that code, by the way. A default tolerance makes more sense for relative/ulp-based comparison.

Which epsilon do you mean, exactly? The absolute difference for comparing near-zero numbers that Bruce Dawson advocates in his blog, or the tolerance used for comparison with relative error or...

I must admit I'm somewhat surprised by being tossed into this discussion after what I believed was idle chatting on Discord. I must say that I'm not particularly invested in...

Hi @neverhood311, that sounds great! Sorry, I haven't meant to ignore your request, I've just been very busy (both at work and in personal life), and I don't have a...

> I had a little time to look into this issue. It's really hard to find details of the implementation anywhere - but I checked out the netlib/eispack code. Their...

By the way, I'll just leave this here as it's something we should consider. Currently, the SVD implementation makes several tests against `T::min_positive_value()`. As you stated in #46, it's preferable...

@AtheMathmo: The stopping criterion was also "fixed" in #46, in the sense that it uses `3 * eps` where `eps` is machine epsilon instead of `T::min_positive_value()`.

@AtheMathmo: I think with decompositions, you don't need to write the algorithm twice. I can think of two alternatives. In both cases, the decomposition algorithm is written exactly once, for...