<@U02D3KPB4JX> We've started a commercial project ...

# datascience

[altavir]

@Adrian Trapletti We've started a commercial project that uses Clarabel4J as one of its solvers. So thank you very much both for it and for your great tutorials. The project itself is closed source, but I think that I will be able to share some code of matrix preparations in KMath examples in future.

Today I made some JMH benchmarks comparing Clarabel solver and OJalgo solver on the same problem. I remember You said that Clarabel should be faster, but I get 1349 ops/second on Ojalgo versus 400 ops/second on Clarabel4J. It is possible that I messed matrix preparations (I did not optimize them), but they all are linear so I do not think it should matter.

[Adrian Trapletti]

Thx for the feedback. Maybe you can share a little example?

[altavir]

I deidentify will, but probably in a few weeks. I also need to check which part I can opensourece. Bat basically I took your example from portfolio planning and did a similar one on Ojalgo.

[altavir]

@Adrian Trapletti An update, we got a larger dataset with about 100 different instruments and Clarabel4J is indeed much faster on it. Ojalgo is about 3 times faster on 5 instruments, but 30 times slower on 100 instruments.

[Adrian Trapletti]

Sounds reasonable to me. For smaller problems the overhead of calling native from Java dominates. For larger problems the solver speed matters.