Do you have a link on this? It's a subject for dat...

# datascience

[gaetan]

Do you have a link on this?
It's a subject for data2viz as we'll probably use a lot of matrix operations for the composition of transformations.

[altavir]

I did not perform native tests myself, but there was a discussion few weeks back about matrix operations in russian kotlin chat an guys used native blas and compared it to EJML on JDK 11. The numbers are very close. I personally made comparison only with python. Let me check if I saved it somewhere

[gaetan]

Thanks, it's enought information.

[altavir]

ojalgo state that it has the same performance for small matrices and better for large ones.

[gaetan]

For us, it's only 2D transformation matrices, so small ones.

[altavir]

you mean 3x3?

[gaetan]

Yes. Could be 4x4 for 3D but it's not in the scope for the moment.

[altavir]

For those you do not have to use any math library at all.

[gaetan]

Yes, I know I can do the math and, because it's a simplified matrix, I can optimize the speed. But I want to make some perf comparison between my code and some libraries.

[altavir]

The problem with small matrices is that memory allocation will be more expensive than actual calculations. You need to optimize different thing.

[gaetan]

Ok. I also need to compare a common code implementation vs some optimized platform implementations.

[altavir]

You can use ojalgo tests for estimation: https://github.com/optimatika/ojAlgo/wiki/Invert For small matrices you should use ejml