I want to use nd4j for production level linear algebra calculations using the many features of ND4J.
- is ND4J still active? I saw some comments that the team left, is it still “safe” for production level?
- my matrices are about 5000x5000 and I do calculations such as element-wise plus,div,cos,sin and more advanced stuff like flattening, cumsum, etc
Is ND4J the best java library to do that? Is there anything better for my use case?
@Tal yes it is! I actively support commercial customers using it at large scale. All of those operations are already in the library. Everything happens in native c++ and we support cpu/gpu. A big release is coming up that will fix a lot of issues with the underlying architecture as well.
That’s great to hear!
Is there a roadmap anywhere where I can see upcoming features and expected release dates?
Also, is there a way to calculate fft using the library?
Thank you for the fast answer
@Tal we use to do ffts via complex numbers but our complex support wasn’t comprehensive enough so it was removed rather than left half done. Due to the ML focus of the library we mainly invested resources in implementing more important things like fp16 and the like.
No roadmap at present as I’m focused on getting a stable 1.0 out with all the basic things fixed like performance, technical debt.
Most of that has been done and I’m just trying to get cuda finalized then will do an M3 release. After M3 I’ll evaluate left over issues and do a 1.0.
In terms of releases, due to the technical debt clean up I"m only doing 1 off releases for customers who need specific functionality but will hopefully have M3 out soon,.
Is there any expected release time for M3 yet?