Using native blas

[lau...@apstat.com]

Hi,

Hope that Mikio can help me out ;)

I tried using jblas to invert a large sparse square matrix (n=3500).

With Colt runtime is around 15 seconds (no blas)
With Numpy runtime is around 7 seconds (anaconda openblas)
With R runtime is around 2 seconds (my system's blas I believe).

When I use jblas, the inversion time is 2 minutes.

When I run java -server -jar jblas-1.2.4.jar, it tells me

org.jblas CONFIG BLAS native library not found in path. Copying native library from the archive.

I tried to set the LD_LIBRARY_PATH to "/usr/lib/libblas/", where there are these files:
/usr/lib/libblas/libblas.a
/usr/lib/libblas/libblas.so
/usr/lib/libblas/libblas.so.3
/usr/lib/libblas/libblas.so.3.0

I still get org.jblas CONFIG BLAS native library not found in path.

I suspect that jblas is being much slower than R and python and Colt because the library from the archive is not well tuned with my system, is that possible? How can Colt be faster than BLAS?

Thanks!
Eric

[Mikio Braun]

Hi Eric,

It may be because jblas doesnt have support for sparse matrices.

Concerning the log messages, in the end it says "copying libs from the archive" which means that it copies the bundled shared libs, which is totally ok behavior.

Apart from that, any other properties on the matrix like it being symmetric?

-M

---

Von: lau...@apstat.com
Gesendet: ‎17.‎09.‎2015 22:32
An: jblas-users
Betreff: Using native blas

--
You received this message because you are subscribed to the Google Groups "jblas-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jblas-users...@googlegroups.com.
To post to this group, send email to jblas...@googlegroups.com.
Visit this group at http://groups.google.com/group/jblas-users.
For more options, visit https://groups.google.com/d/optout.

[lau...@apstat.com]

Thanks for the quick response!
It it symmetric in shape but not the same values I believe, I would have to check.

I was saying it is sparse not because I want sparse format, just for more information :)

Shouldnt jblas be as fast as numpy/blas for inversion?