NumNet Congrats!

Very nice library!

Have you thought about including Benchmarks with BenchmarkDotNet? I'm interested how the performance compares with python numpy and NumSharp on Github.

Dec 06 '18 13:12 fdncred

Hello,

Thanks for your message, I didn't knew about NumSharp before.

I added a new branch to the repo with a few benchmarks to compare NumNet to NumSharp. From what I tested, NumNet compares quite favorably especially when using MKL for the base operations (for instance, Dot function can be up to 100x faster in NumNet). Without MKL, NumNet is on average 2 times faster for most base functions that I tried.

If you want to add some more benchmarks or have any track of improvement for NumNet don't hesitate to let us know, we would be happy to collaborate with the C# community to improve our library.

Dec 11 '18 16:12 thp-proxem

Thanks for the response. We're working now on adding LAPACK for speed ups.

Dec 11 '18 16:12 fdncred

I'm also looking forward to check the performance comparison. I think MKL will be the winner because it optimized by Intel, perfect running on multiple core. @fdncred We're also considering support multiple provider solution, LAPACK and MKL.

Dec 12 '18 19:12 Oceania2018

@Oceania2018, Ya, that's why I liked Armadillo. It seemed to support multiple libraries. Last I looked it seemed like OpenBlas was the winner across platforms and architecture.

@thp-proxem, Did you try other linear algebra/maths libraries before choosing MKL? If so, what were they and what pushed to to choose MKL over the others?

Dec 12 '18 20:12 fdncred