Dr. Juno Woods
Dr. Juno Woods
It's been a really long time since I looked at the source code for Yale, unfortunately. You could look at the changelog for that file and try to figure out...
How would you go about fixing it?
What happens when elements get deleted? What happens when the matrix's non-diagonal default is non-zero?
It seems like `ndnz` is defined as being part of the IJA array, isn't it? So have we duplicated this by including it in the Yale struct also? Argh, frustrating...
I suspect one reason I didn't remove it before is that it causes the data in yale, list, and dense to be in the same relative memory location.
Yes — the problem being that ATLAS doesn't support quads. We could do this, but it would need to be part of eliminating the unnecessary rational types. Why only for...
I'm not putting a milestone on this enhancement because I think it needs some more design work before it can really be considered achievable.
Eww, that is painfully slow. However, I'd like — if possible — a little more information on _why_ it's slow. I notice that `mean` calls `inject_rank`: https://github.com/SciRuby/nmatrix/blob/721942cc865f01c3d1092868c8aeb6e4bed40466/lib/nmatrix/enumerate.rb#L217 I notice that...
Well, I think there's an easier way to do what you did in that fix. Keep in mind that each_with_indices is going to call dense_map also, so you're adding an...
Oh, true. Well, in that case, I'm having trouble figuring out why your modification should be faster than what was there before. It seems like you're doing an extra copy....