Install OpenBLAS on Anaconda numpy on linux

[Parkway]

Has anyone successfully installed OpenBlas on a Anaconda python distribution with numpy on linux - ubuntu or centos?

[Zhang Xianyi]

Sorry. I didn't test this case before. What's the problem?

Xianyi

2015-06-21 9:49 GMT-05:00 Parkway <dinesh...@hotmail.com>:

Has anyone successfully installed OpenBlas on a Anaconda python distribution with numpy on linux - ubuntu or centos?

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

[Olivier Grisel]

You would need to rebuild numpy and scipy from source which defeats
the purpose of using anaconda in the first place in my opinion.

--
Olivier

[Parkway]

Thank-you!  Anaconda + numpy for Win64 (w/ MKL) and OS/X (w/ Accelerate) are fine. But, for Linux w/ Atlas is a problem. One one multi-core machine it recognizes the 4-cores but on a 16-core machine it doesn't. I would like to standardize on OpenBlas but the option with Anaconda doesn't exist.

Anaconda offers a lot of benefits over re-building from source.  However, I think the Anaconda folks should offer alternative BLAS binary libraries such as OpenBlas.

[Olivier Grisel]

The Anaconda developers sell the MKL optimizations as an addon:

https://store.continuum.io/cshop/mkl-optimizations/

So they probably won't have the incentives to ship OpenBLAS as a free
alternative with similar performance. But you can always contact them
to ask.

That being said you should be able to rebuild numpy and scipy from
source provided that you have gcc / g++ / gfotran installed on your
machine and that you put the following files renamed as "site.cfg"
inside the numpy and scipy source folders respectively:

https://github.com/ogrisel/docker-sklearn-openblas/blob/master/numpy-site.cfg
https://github.com/ogrisel/docker-sklearn-openblas/blob/master/scipy-site.cfg

If you have issues you should problably report problems on
stackoverflow with detailed error messages and the numpy / scipy /
openblas tags.

--
Olivier

[Parkway]

I'll drop a note on the Anaconda list but like you say it is probably not in their interest though it really should be.

Wanted to avoid going back to building numpy and scipy from source - been there, done it and got the scars! But, maybe that is the only way.

Btw, I found this conda recipe for installing numpy with openblas - https://binstar.org/jakirkham/openblas.