Use native dense-linear-algebra instead of HMatrix

idontgetoutmuch / random-fu-multivariate

Multivariate distributions based on random-fu

BSD 3-Clause "New" or "Revised" License

7 stars 5 forks source link

Use native dense-linear-algebra instead of HMatrix #2

Open ocramz opened 6 years ago

ocramz commented 6 years ago

I've submitted a PR to statistics that extracts the algebra engine as a separate library; which provides a QR factorization and the usual BLAS stuff. Hopefully it will be merged and published soon.

https://github.com/bos/statistics/pull/143

idontgetoutmuch commented 6 years ago

How would moving to this affect performance?

ocramz commented 6 years ago

I'm not sure (yet), but what I can say is that it would make the package more portable. The statistics algebra engine uses mutable vectors and has been optimized quite a lot as far as I can tell.